QwenImage

Qwen-Image from the Qwen team is an image generation foundation model in the Qwen series that achieves significant advances in complex text rendering and precise image editing. Experiments show strong general capabilities in both image generation and editing, with exceptional performance in text rendering, especially for Chinese.

Qwen-Image comes in the following variants:

model type	model id
Qwen-Image	Qwen/Qwen-Image
Qwen-Image-Edit	Qwen/Qwen-Image-Edit

Tip

Make sure to check out the Schedulers guide to learn how to explore the tradeoff between scheduler speed and quality, and see the reuse components across pipelines section to learn how to efficiently load the same components into multiple pipelines.

LoRA for faster inference

Use a LoRA from lightx2v/Qwen-Image-Lightning to speed up inference by reducing the number of steps. Refer to the code snippet below:

from mindone.diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler
import mindspore
import math

scheduler_config = {
    "base_image_seq_len": 256,
    "base_shift": math.log(3),  # We use shift=3 in distillation
    "invert_sigmas": False,
    "max_image_seq_len": 8192,
    "max_shift": math.log(3),  # We use shift=3 in distillation
    "num_train_timesteps": 1000,
    "shift": 1.0,
    "shift_terminal": None,  # set shift_terminal to None
    "stochastic_sampling": False,
    "time_shift_type": "exponential",
    "use_beta_sigmas": False,
    "use_dynamic_shifting": True,
    "use_exponential_sigmas": False,
    "use_karras_sigmas": False,
}
scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)
pipe = DiffusionPipeline.from_pretrained(
    "Qwen/Qwen-Image", scheduler=scheduler, mindspore_dtype=mindspore.bfloat16
)
pipe.load_lora_weights(
    "Qwen/lightx2v/Qwen-Image-Lightning",
    weight_name="Qwen-Image-Lightning-8steps-V1.0.safetensors",
    adapter_name="qwenimage-lora"
)
pipe.fuse_lora()
pipe.unload_lora_weights()

prompt = "a tiny astronaut hatching from an egg on the moon, Ultra HD, 4K, cinematic composition."
negative_prompt = " "
image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=1024,
    height=1024,
    num_inference_steps=8,
    true_cfg_scale=1.0,
    generator=None,
)[0][0]
image.save("lora_pic/qwen_fewsteps_lora.png")

Tip

The guidance_scale parameter in the pipeline is there to support future guidance-distilled models when they come up. Note that passing guidance_scale to the pipeline is ineffective. To enable classifier-free guidance, please pass true_cfg_scale and negative_prompt (even an empty negative prompt like " ") should enable classifier-free guidance computations.

mindone.diffusers.QwenImagePipeline

Bases: DiffusionPipeline, QwenImageLoraLoaderMixin

The QwenImage pipeline for text-to-image generation.

PARAMETER	DESCRIPTION
transformer	Conditional Transformer (MMDiT) architecture to denoise the encoded image latents. TYPE: [`QwenImageTransformer2DModel`]
scheduler	A scheduler to be used in combination with transformer to denoise the encoded image latents. TYPE: [`FlowMatchEulerDiscreteScheduler`]
vae	Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. TYPE: [`AutoencoderKL`]
text_encoder	Qwen2.5-VL-7B-Instruct, specifically the Qwen2.5-VL-7B-Instruct variant. TYPE: [`Qwen2.5-VL-7B-Instruct`]
tokenizer	Tokenizer of class CLIPTokenizer. TYPE: `QwenTokenizer`

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage.py

class QwenImagePipeline(DiffusionPipeline, QwenImageLoraLoaderMixin):
    r"""
    The QwenImage pipeline for text-to-image generation.

    Args:
        transformer ([`QwenImageTransformer2DModel`]):
            Conditional Transformer (MMDiT) architecture to denoise the encoded image latents.
        scheduler ([`FlowMatchEulerDiscreteScheduler`]):
            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
        vae ([`AutoencoderKL`]):
            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
        text_encoder ([`Qwen2.5-VL-7B-Instruct`]):
            [Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct), specifically the
            [Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct) variant.
        tokenizer (`QwenTokenizer`):
            Tokenizer of class
            [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer).
    """

    model_cpu_offload_seq = "text_encoder->transformer->vae"
    _callback_tensor_inputs = ["latents", "prompt_embeds"]

    def __init__(
        self,
        scheduler: FlowMatchEulerDiscreteScheduler,
        vae: AutoencoderKLQwenImage,
        text_encoder: Qwen2_5_VLForConditionalGeneration,
        tokenizer: Qwen2Tokenizer,
        transformer: QwenImageTransformer2DModel,
    ):
        super().__init__()

        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            tokenizer=tokenizer,
            transformer=transformer,
            scheduler=scheduler,
        )
        self.vae_scale_factor = 2 ** len(self.vae.temperal_downsample) if getattr(self, "vae", None) else 8
        # QwenImage latents are turned into 2x2 patches and packed. This means the latent width and height has to be divisible
        # by the patch size. So the vae scale factor is multiplied by the patch size to account for this
        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor * 2)
        self.tokenizer_max_length = 1024
        self.prompt_template_encode = (
            "<|im_start|>system\nDescribe the image by detailing the color, shape, size, "
            "texture, quantity, text, spatial relationships of the objects and background:"
            "<|im_end|>\n<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
        )
        self.prompt_template_encode_start_idx = 34
        self.default_sample_size = 128

    def _extract_masked_hidden(self, hidden_states: ms.Tensor, mask: ms.Tensor):
        bool_mask = mask.bool()
        valid_lengths = bool_mask.sum(dim=1)
        selected = hidden_states[bool_mask]
        split_result = mint.split(selected, valid_lengths.tolist(), dim=0)

        return split_result

    def _get_qwen_prompt_embeds(
        self,
        prompt: Union[str, List[str]] = None,
        dtype: Optional[ms.Type] = None,
    ):
        dtype = dtype or self.text_encoder.dtype

        prompt = [prompt] if isinstance(prompt, str) else prompt

        template = self.prompt_template_encode
        drop_idx = self.prompt_template_encode_start_idx
        txt = [template.format(e) for e in prompt]
        txt_tokens = self.tokenizer(
            txt, max_length=self.tokenizer_max_length + drop_idx, padding=True, truncation=True, return_tensors="np"
        )
        encoder_hidden_states = self.text_encoder(
            input_ids=ms.tensor(txt_tokens.input_ids),
            attention_mask=ms.tensor(txt_tokens.attention_mask),
            output_hidden_states=True,
        )
        hidden_states = encoder_hidden_states.hidden_states[-1]
        split_hidden_states = self._extract_masked_hidden(hidden_states, ms.tensor(txt_tokens.attention_mask))
        split_hidden_states = [e[drop_idx:] for e in split_hidden_states]
        attn_mask_list = [mint.ones(e.shape[0], dtype=ms.int64) for e in split_hidden_states]
        max_seq_len = max([e.shape[0] for e in split_hidden_states])
        prompt_embeds = mint.stack(
            [mint.cat([u, u.new_zeros((max_seq_len - u.shape[0], u.shape[1]))]) for u in split_hidden_states]
        )
        encoder_attention_mask = mint.stack(
            [mint.cat([u, u.new_zeros((max_seq_len - u.shape[0]))]) for u in attn_mask_list]
        )

        prompt_embeds = prompt_embeds.to(dtype=dtype)

        return prompt_embeds, encoder_attention_mask

    def encode_prompt(
        self,
        prompt: Union[str, List[str]],
        num_images_per_prompt: int = 1,
        prompt_embeds: Optional[ms.Tensor] = None,
        prompt_embeds_mask: Optional[ms.Tensor] = None,
        max_sequence_length: int = 1024,
    ):
        r"""

        Args:
            prompt (`str` or `List[str]`, *optional*):
                prompt to be encoded
            num_images_per_prompt (`int`):
                number of images that should be generated per prompt
            prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
        """
        prompt = [prompt] if isinstance(prompt, str) else prompt
        batch_size = len(prompt) if prompt_embeds is None else prompt_embeds.shape[0]

        if prompt_embeds is None:
            prompt_embeds, prompt_embeds_mask = self._get_qwen_prompt_embeds(prompt)

        prompt_embeds = prompt_embeds[:, :max_sequence_length]
        prompt_embeds_mask = prompt_embeds_mask[:, :max_sequence_length]

        _, seq_len, _ = prompt_embeds.shape
        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
        prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
        prompt_embeds_mask = prompt_embeds_mask.repeat(1, num_images_per_prompt, 1)
        prompt_embeds_mask = prompt_embeds_mask.view(batch_size * num_images_per_prompt, seq_len)

        return prompt_embeds, prompt_embeds_mask

    def check_inputs(
        self,
        prompt,
        height,
        width,
        negative_prompt=None,
        prompt_embeds=None,
        negative_prompt_embeds=None,
        prompt_embeds_mask=None,
        negative_prompt_embeds_mask=None,
        callback_on_step_end_tensor_inputs=None,
        max_sequence_length=None,
    ):
        if height % (self.vae_scale_factor * 2) != 0 or width % (self.vae_scale_factor * 2) != 0:
            logger.warning(
                f"`height` and `width` have to be divisible by {self.vae_scale_factor * 2} but are {height} and {width}."
                " Dimensions will be resized accordingly"
            )

        if callback_on_step_end_tensor_inputs is not None and not all(
            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
        ):
            raise ValueError(
                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found"
                f" {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
            )

        if prompt is not None and prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt is None and prompt_embeds is None:
            raise ValueError(
                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
            )
        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")

        if negative_prompt is not None and negative_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
            )

        if prompt_embeds is not None and prompt_embeds_mask is None:
            raise ValueError(
                "If `prompt_embeds` are provided, `prompt_embeds_mask` also have to be passed. Make sure to generate `prompt_embeds_mask`"
                " from the same text encoder that was used to generate `prompt_embeds`."
            )

        if negative_prompt_embeds is not None and negative_prompt_embeds_mask is None:
            raise ValueError(
                "If `negative_prompt_embeds` are provided, `negative_prompt_embeds_mask` also have to be passed. Make sure to generate"
                " `negative_prompt_embeds_mask` from the same text encoder that was used to generate `negative_prompt_embeds`."
            )

        if max_sequence_length is not None and max_sequence_length > 1024:
            raise ValueError(f"`max_sequence_length` cannot be greater than 1024 but is {max_sequence_length}")

    @staticmethod
    def _pack_latents(latents, batch_size, num_channels_latents, height, width):
        latents = latents.view(batch_size, num_channels_latents, height // 2, 2, width // 2, 2)
        latents = latents.permute(0, 2, 4, 1, 3, 5)
        latents = latents.reshape(batch_size, (height // 2) * (width // 2), num_channels_latents * 4)

        return latents

    @staticmethod
    def _unpack_latents(latents, height, width, vae_scale_factor):
        batch_size, num_patches, channels = latents.shape

        # VAE applies 8x compression on images but we must also account for packing which requires
        # latent height and width to be divisible by 2.
        height = 2 * (int(height) // (vae_scale_factor * 2))
        width = 2 * (int(width) // (vae_scale_factor * 2))

        latents = latents.view(batch_size, height // 2, width // 2, channels // 4, 2, 2)
        latents = latents.permute(0, 3, 1, 4, 2, 5)

        latents = latents.reshape(batch_size, channels // (2 * 2), 1, height, width)

        return latents

    def enable_vae_slicing(self):
        r"""
        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
        """
        self.vae.enable_slicing()

    def disable_vae_slicing(self):
        r"""
        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
        computing decoding in one step.
        """
        self.vae.disable_slicing()

    def enable_vae_tiling(self):
        r"""
        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
        processing larger images.
        """
        self.vae.enable_tiling()

    def disable_vae_tiling(self):
        r"""
        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
        computing decoding in one step.
        """
        self.vae.disable_tiling()

    def prepare_latents(
        self,
        batch_size,
        num_channels_latents,
        height,
        width,
        dtype,
        generator,
        latents=None,
    ):
        # VAE applies 8x compression on images but we must also account for packing which requires
        # latent height and width to be divisible by 2.
        height = 2 * (int(height) // (self.vae_scale_factor * 2))
        width = 2 * (int(width) // (self.vae_scale_factor * 2))

        shape = (batch_size, 1, num_channels_latents, height, width)

        if latents is not None:
            return latents.to(dtype=dtype)

        if isinstance(generator, list) and len(generator) != batch_size:
            raise ValueError(
                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
            )

        latents = randn_tensor(shape, generator=generator, dtype=dtype)
        latents = self._pack_latents(latents, batch_size, num_channels_latents, height, width)

        return latents

    @property
    def guidance_scale(self):
        return self._guidance_scale

    @property
    def attention_kwargs(self):
        return self._attention_kwargs

    @property
    def num_timesteps(self):
        return self._num_timesteps

    @property
    def current_timestep(self):
        return self._current_timestep

    @property
    def interrupt(self):
        return self._interrupt

    def __call__(
        self,
        prompt: Union[str, List[str]] = None,
        negative_prompt: Union[str, List[str]] = None,
        true_cfg_scale: float = 4.0,
        height: Optional[int] = None,
        width: Optional[int] = None,
        num_inference_steps: int = 50,
        sigmas: Optional[List[float]] = None,
        guidance_scale: Optional[float] = None,
        num_images_per_prompt: int = 1,
        generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
        latents: Optional[ms.Tensor] = None,
        prompt_embeds: Optional[ms.Tensor] = None,
        prompt_embeds_mask: Optional[ms.Tensor] = None,
        negative_prompt_embeds: Optional[ms.Tensor] = None,
        negative_prompt_embeds_mask: Optional[ms.Tensor] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = True,
        attention_kwargs: Optional[Dict[str, Any]] = None,
        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
        max_sequence_length: int = 512,
    ):
        r"""
        Function invoked when calling the pipeline for generation.

        Args:
            prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
                instead.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation. If not defined, one has to pass
                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
                not greater than `1`).
            true_cfg_scale (`float`, *optional*, defaults to 1.0):
                Guidance scale as defined in [Classifier-Free Diffusion
                Guidance](https://huggingface.co/papers/2207.12598). `true_cfg_scale` is defined as `w` of equation 2.
                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Classifier-free guidance is enabled by
                setting `true_cfg_scale > 1` and a provided `negative_prompt`. Higher guidance scale encourages to
                generate images that are closely linked to the text `prompt`, usually at the expense of lower image
                quality.
            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                The height in pixels of the generated image. This is set to 1024 by default for the best results.
            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                The width in pixels of the generated image. This is set to 1024 by default for the best results.
            num_inference_steps (`int`, *optional*, defaults to 50):
                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                expense of slower inference.
            sigmas (`List[float]`, *optional*):
                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                will be used.
            guidance_scale (`float`, *optional*, defaults to None):
                A guidance scale value for guidance distilled models. Unlike the traditional classifier-free guidance
                where the guidance scale is applied during inference through noise prediction rescaling, guidance
                distilled models take the guidance scale directly as an input parameter during forward pass. Guidance
                scale is enabled by setting `guidance_scale > 1`. Higher guidance scale encourages to generate images
                that are closely linked to the text `prompt`, usually at the expense of lower image quality. This
                parameter in the pipeline is there to support future guidance-distilled models when they come up. It is
                ignored when not using guidance distilled models. To enable traditional classifier-free guidance,
                please pass `true_cfg_scale > 1.0` and `negative_prompt` (even an empty negative prompt like " " should
                enable classifier-free guidance computations).
            num_images_per_prompt (`int`, *optional*, defaults to 1):
                The number of images to generate per prompt.
            generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
                One or a list of [numpy generator(s)](https://numpy.org/doc/stable/reference/random/generator.html)
                to make generation deterministic.
            latents (`ms.Tensor`, *optional*):
                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor will be generated by sampling using the supplied random `generator`.
            prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
            negative_prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                argument.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generate image. Choose between
                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.qwenimage.QwenImagePipelineOutput`] instead of a plain tuple.
            attention_kwargs (`dict`, *optional*):
                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
                `self.processor` in
                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
            callback_on_step_end (`Callable`, *optional*):
                A function that calls at the end of each denoising steps during the inference. The function is called
                with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
                callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
                `callback_on_step_end_tensor_inputs`.
            callback_on_step_end_tensor_inputs (`List`, *optional*):
                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
                `._callback_tensor_inputs` attribute of your pipeline class.
            max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.

        Examples:

        Returns:
            [`~pipelines.qwenimage.QwenImagePipelineOutput`] or `tuple`:
            [`~pipelines.qwenimage.QwenImagePipelineOutput`] if `return_dict` is True, otherwise a `tuple`. When
            returning a tuple, the first element is a list with the generated images.
        """

        height = height or self.default_sample_size * self.vae_scale_factor
        width = width or self.default_sample_size * self.vae_scale_factor

        # 1. Check inputs. Raise error if not correct
        self.check_inputs(
            prompt,
            height,
            width,
            negative_prompt=negative_prompt,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            prompt_embeds_mask=prompt_embeds_mask,
            negative_prompt_embeds_mask=negative_prompt_embeds_mask,
            callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
            max_sequence_length=max_sequence_length,
        )

        self._guidance_scale = guidance_scale
        self._attention_kwargs = attention_kwargs
        self._current_timestep = None
        self._interrupt = False

        # 2. Define call parameters
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        has_neg_prompt = negative_prompt is not None or (
            negative_prompt_embeds is not None and negative_prompt_embeds_mask is not None
        )

        if true_cfg_scale > 1 and not has_neg_prompt:
            logger.warning(
                f"true_cfg_scale is passed as {true_cfg_scale}, but classifier-free guidance is not enabled since no negative_prompt is provided."
            )
        elif true_cfg_scale <= 1 and has_neg_prompt:
            logger.warning(
                " negative_prompt is passed but classifier-free guidance is not enabled since true_cfg_scale <= 1"
            )

        do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
        prompt_embeds, prompt_embeds_mask = self.encode_prompt(
            prompt=prompt,
            prompt_embeds=prompt_embeds,
            prompt_embeds_mask=prompt_embeds_mask,
            num_images_per_prompt=num_images_per_prompt,
            max_sequence_length=max_sequence_length,
        )
        if do_true_cfg:
            negative_prompt_embeds, negative_prompt_embeds_mask = self.encode_prompt(
                prompt=negative_prompt,
                prompt_embeds=negative_prompt_embeds,
                prompt_embeds_mask=negative_prompt_embeds_mask,
                num_images_per_prompt=num_images_per_prompt,
                max_sequence_length=max_sequence_length,
            )

        # 4. Prepare latent variables
        num_channels_latents = self.transformer.config.in_channels // 4
        latents = self.prepare_latents(
            batch_size * num_images_per_prompt,
            num_channels_latents,
            height,
            width,
            prompt_embeds.dtype,
            generator,
            latents,
        )
        img_shapes = [[(1, height // self.vae_scale_factor // 2, width // self.vae_scale_factor // 2)]] * batch_size

        # 5. Prepare timesteps
        sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else sigmas
        image_seq_len = latents.shape[1]
        mu = calculate_shift(
            image_seq_len,
            self.scheduler.config.get("base_image_seq_len", 256),
            self.scheduler.config.get("max_image_seq_len", 4096),
            self.scheduler.config.get("base_shift", 0.5),
            self.scheduler.config.get("max_shift", 1.15),
        )
        timesteps, num_inference_steps = retrieve_timesteps(
            self.scheduler,
            num_inference_steps,
            sigmas=sigmas,
            mu=mu,
        )
        num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
        self._num_timesteps = len(timesteps)

        # handle guidance
        if self.transformer.config.guidance_embeds and guidance_scale is None:
            raise ValueError("guidance_scale is required for guidance-distilled model.")
        elif self.transformer.config.guidance_embeds:
            guidance = mint.full([1], guidance_scale, dtype=ms.float32)
            guidance = guidance.expand((latents.shape[0],))
        elif not self.transformer.config.guidance_embeds and guidance_scale is not None:
            logger.warning(
                f"guidance_scale is passed as {guidance_scale}, but ignored since the model is not guidance-distilled."
            )
            guidance = None
        elif not self.transformer.config.guidance_embeds and guidance_scale is None:
            guidance = None

        if self.attention_kwargs is None:
            self._attention_kwargs = {}

        txt_seq_lens = prompt_embeds_mask.sum(dim=1).tolist() if prompt_embeds_mask is not None else None
        negative_txt_seq_lens = (
            negative_prompt_embeds_mask.sum(dim=1).tolist() if negative_prompt_embeds_mask is not None else None
        )

        # 6. Denoising loop
        self.scheduler.set_begin_index(0)
        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(timesteps):
                if self.interrupt:
                    continue

                self._current_timestep = t
                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
                timestep = t.expand((latents.shape[0],)).to(latents.dtype)
                noise_pred = self.transformer(
                    hidden_states=latents,
                    timestep=timestep / 1000,
                    guidance=guidance,
                    encoder_hidden_states_mask=prompt_embeds_mask,
                    encoder_hidden_states=prompt_embeds,
                    img_shapes=img_shapes,
                    txt_seq_lens=txt_seq_lens,
                    attention_kwargs=self.attention_kwargs,
                    return_dict=False,
                )[0]

                if do_true_cfg:
                    neg_noise_pred = self.transformer(
                        hidden_states=latents,
                        timestep=timestep / 1000,
                        guidance=guidance,
                        encoder_hidden_states_mask=negative_prompt_embeds_mask,
                        encoder_hidden_states=negative_prompt_embeds,
                        img_shapes=img_shapes,
                        txt_seq_lens=negative_txt_seq_lens,
                        attention_kwargs=self.attention_kwargs,
                        return_dict=False,
                    )[0]
                    comb_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)

                    cond_norm = mint.norm(noise_pred, dim=-1, keepdim=True)
                    noise_norm = mint.norm(comb_pred, dim=-1, keepdim=True)
                    noise_pred = comb_pred * (cond_norm / noise_norm)

                # compute the previous noisy sample x_t -> x_t-1
                latents_dtype = latents.dtype
                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]

                if latents.dtype != latents_dtype:
                    latents = latents.to(latents_dtype)

                if callback_on_step_end is not None:
                    callback_kwargs = {}
                    for k in callback_on_step_end_tensor_inputs:
                        callback_kwargs[k] = locals()[k]
                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)

                    latents = callback_outputs.pop("latents", latents)
                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)

                # call the callback, if provided
                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                    progress_bar.update()

        self._current_timestep = None
        if output_type == "latent":
            image = latents
        else:
            latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
            latents = latents.to(self.vae.dtype)
            latents_mean = (
                ms.tensor(self.vae.config.latents_mean).view(1, self.vae.config.z_dim, 1, 1, 1).to(latents.dtype)
            )
            latents_std = 1.0 / ms.tensor(self.vae.config.latents_std).view(1, self.vae.config.z_dim, 1, 1, 1).to(
                latents.dtype
            )
            latents = latents / latents_std + latents_mean
            image = self.vae.decode(latents, return_dict=False)[0][:, :, 0]
            image = self.image_processor.postprocess(image, output_type=output_type)

        if not return_dict:
            return (image,)

        return QwenImagePipelineOutput(images=image)

mindone.diffusers.QwenImagePipeline.__call__(prompt=None, negative_prompt=None, true_cfg_scale=4.0, height=None, width=None, num_inference_steps=50, sigmas=None, guidance_scale=None, num_images_per_prompt=1, generator=None, latents=None, prompt_embeds=None, prompt_embeds_mask=None, negative_prompt_embeds=None, negative_prompt_embeds_mask=None, output_type='pil', return_dict=True, attention_kwargs=None, callback_on_step_end=None, callback_on_step_end_tensor_inputs=['latents'], max_sequence_length=512)

Function invoked when calling the pipeline for generation.

PARAMETER	DESCRIPTION
prompt	The prompt or prompts to guide the image generation. If not defined, one has to pass prompt_embeds. instead. TYPE: `str` or `List[str]`, *optional* DEFAULT: None
negative_prompt	The prompt or prompts not to guide the image generation. If not defined, one has to pass negative_prompt_embeds instead. Ignored when not using guidance (i.e., ignored if true_cfg_scale is not greater than 1). TYPE: `str` or `List[str]`, *optional* DEFAULT: None
true_cfg_scale	Guidance scale as defined in Classifier-Free Diffusion Guidance. true_cfg_scale is defined as w of equation 2. of Imagen Paper. Classifier-free guidance is enabled by setting true_cfg_scale > 1 and a provided negative_prompt. Higher guidance scale encourages to generate images that are closely linked to the text prompt, usually at the expense of lower image quality. TYPE: `float`, *optional*, defaults to 1.0 DEFAULT: 4.0
height	The height in pixels of the generated image. This is set to 1024 by default for the best results. TYPE: `int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor DEFAULT: None
width	The width in pixels of the generated image. This is set to 1024 by default for the best results. TYPE: `int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor DEFAULT: None
num_inference_steps	The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. TYPE: `int`, *optional*, defaults to 50 DEFAULT: 50
sigmas	Custom sigmas to use for the denoising process with schedulers which support a sigmas argument in their set_timesteps method. If not defined, the default behavior when num_inference_steps is passed will be used. TYPE: `List[float]`, *optional* DEFAULT: None
guidance_scale	A guidance scale value for guidance distilled models. Unlike the traditional classifier-free guidance where the guidance scale is applied during inference through noise prediction rescaling, guidance distilled models take the guidance scale directly as an input parameter during forward pass. Guidance scale is enabled by setting guidance_scale > 1. Higher guidance scale encourages to generate images that are closely linked to the text prompt, usually at the expense of lower image quality. This parameter in the pipeline is there to support future guidance-distilled models when they come up. It is ignored when not using guidance distilled models. To enable traditional classifier-free guidance, please pass true_cfg_scale > 1.0 and negative_prompt (even an empty negative prompt like " " should enable classifier-free guidance computations). TYPE: `float`, *optional*, defaults to None DEFAULT: None
num_images_per_prompt	The number of images to generate per prompt. TYPE: `int`, *optional*, defaults to 1 DEFAULT: 1
generator	One or a list of numpy generator(s) to make generation deterministic. TYPE: `np.random.Generator` or `List[np.random.Generator]`, *optional* DEFAULT: None
latents	Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor will be generated by sampling using the supplied random generator. TYPE: `ms.Tensor`, *optional* DEFAULT: None
prompt_embeds	Pre-generated text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, text embeddings will be generated from prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
negative_prompt_embeds	Pre-generated negative text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, negative_prompt_embeds will be generated from negative_prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
output_type	The output format of the generate image. Choose between PIL: PIL.Image.Image or np.array. TYPE: `str`, *optional*, defaults to `"pil"` DEFAULT: 'pil'
return_dict	Whether or not to return a [~pipelines.qwenimage.QwenImagePipelineOutput] instead of a plain tuple. TYPE: `bool`, *optional*, defaults to `True` DEFAULT: True
attention_kwargs	A kwargs dictionary that if specified is passed along to the AttentionProcessor as defined under self.processor in diffusers.models.attention_processor. TYPE: `dict`, *optional* DEFAULT: None
callback_on_step_end	A function that calls at the end of each denoising steps during the inference. The function is called with the following arguments: callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict). callback_kwargs will include a list of all tensors as specified by callback_on_step_end_tensor_inputs. TYPE: `Callable`, *optional* DEFAULT: None
callback_on_step_end_tensor_inputs	The list of tensor inputs for the callback_on_step_end function. The tensors specified in the list will be passed as callback_kwargs argument. You will only be able to include variables listed in the ._callback_tensor_inputs attribute of your pipeline class. TYPE: `List`, *optional* DEFAULT: ['latents']
max_sequence_length	Maximum sequence length to use with the prompt. TYPE: `int` defaults to 512 DEFAULT: 512

RETURNS	DESCRIPTION
	[~pipelines.qwenimage.QwenImagePipelineOutput] or tuple:
	[~pipelines.qwenimage.QwenImagePipelineOutput] if return_dict is True, otherwise a tuple. When
	returning a tuple, the first element is a list with the generated images.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage.py

def __call__(
    self,
    prompt: Union[str, List[str]] = None,
    negative_prompt: Union[str, List[str]] = None,
    true_cfg_scale: float = 4.0,
    height: Optional[int] = None,
    width: Optional[int] = None,
    num_inference_steps: int = 50,
    sigmas: Optional[List[float]] = None,
    guidance_scale: Optional[float] = None,
    num_images_per_prompt: int = 1,
    generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
    latents: Optional[ms.Tensor] = None,
    prompt_embeds: Optional[ms.Tensor] = None,
    prompt_embeds_mask: Optional[ms.Tensor] = None,
    negative_prompt_embeds: Optional[ms.Tensor] = None,
    negative_prompt_embeds_mask: Optional[ms.Tensor] = None,
    output_type: Optional[str] = "pil",
    return_dict: bool = True,
    attention_kwargs: Optional[Dict[str, Any]] = None,
    callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
    callback_on_step_end_tensor_inputs: List[str] = ["latents"],
    max_sequence_length: int = 512,
):
    r"""
    Function invoked when calling the pipeline for generation.

    Args:
        prompt (`str` or `List[str]`, *optional*):
            The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
            instead.
        negative_prompt (`str` or `List[str]`, *optional*):
            The prompt or prompts not to guide the image generation. If not defined, one has to pass
            `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
            not greater than `1`).
        true_cfg_scale (`float`, *optional*, defaults to 1.0):
            Guidance scale as defined in [Classifier-Free Diffusion
            Guidance](https://huggingface.co/papers/2207.12598). `true_cfg_scale` is defined as `w` of equation 2.
            of [Imagen Paper](https://huggingface.co/papers/2205.11487). Classifier-free guidance is enabled by
            setting `true_cfg_scale > 1` and a provided `negative_prompt`. Higher guidance scale encourages to
            generate images that are closely linked to the text `prompt`, usually at the expense of lower image
            quality.
        height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
            The height in pixels of the generated image. This is set to 1024 by default for the best results.
        width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
            The width in pixels of the generated image. This is set to 1024 by default for the best results.
        num_inference_steps (`int`, *optional*, defaults to 50):
            The number of denoising steps. More denoising steps usually lead to a higher quality image at the
            expense of slower inference.
        sigmas (`List[float]`, *optional*):
            Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
            their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
            will be used.
        guidance_scale (`float`, *optional*, defaults to None):
            A guidance scale value for guidance distilled models. Unlike the traditional classifier-free guidance
            where the guidance scale is applied during inference through noise prediction rescaling, guidance
            distilled models take the guidance scale directly as an input parameter during forward pass. Guidance
            scale is enabled by setting `guidance_scale > 1`. Higher guidance scale encourages to generate images
            that are closely linked to the text `prompt`, usually at the expense of lower image quality. This
            parameter in the pipeline is there to support future guidance-distilled models when they come up. It is
            ignored when not using guidance distilled models. To enable traditional classifier-free guidance,
            please pass `true_cfg_scale > 1.0` and `negative_prompt` (even an empty negative prompt like " " should
            enable classifier-free guidance computations).
        num_images_per_prompt (`int`, *optional*, defaults to 1):
            The number of images to generate per prompt.
        generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
            One or a list of [numpy generator(s)](https://numpy.org/doc/stable/reference/random/generator.html)
            to make generation deterministic.
        latents (`ms.Tensor`, *optional*):
            Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
            generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
            tensor will be generated by sampling using the supplied random `generator`.
        prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
            provided, text embeddings will be generated from `prompt` input argument.
        negative_prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
            weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
            argument.
        output_type (`str`, *optional*, defaults to `"pil"`):
            The output format of the generate image. Choose between
            [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
        return_dict (`bool`, *optional*, defaults to `True`):
            Whether or not to return a [`~pipelines.qwenimage.QwenImagePipelineOutput`] instead of a plain tuple.
        attention_kwargs (`dict`, *optional*):
            A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
            `self.processor` in
            [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
        callback_on_step_end (`Callable`, *optional*):
            A function that calls at the end of each denoising steps during the inference. The function is called
            with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
            callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
            `callback_on_step_end_tensor_inputs`.
        callback_on_step_end_tensor_inputs (`List`, *optional*):
            The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
            will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
            `._callback_tensor_inputs` attribute of your pipeline class.
        max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.

    Examples:

    Returns:
        [`~pipelines.qwenimage.QwenImagePipelineOutput`] or `tuple`:
        [`~pipelines.qwenimage.QwenImagePipelineOutput`] if `return_dict` is True, otherwise a `tuple`. When
        returning a tuple, the first element is a list with the generated images.
    """

    height = height or self.default_sample_size * self.vae_scale_factor
    width = width or self.default_sample_size * self.vae_scale_factor

    # 1. Check inputs. Raise error if not correct
    self.check_inputs(
        prompt,
        height,
        width,
        negative_prompt=negative_prompt,
        prompt_embeds=prompt_embeds,
        negative_prompt_embeds=negative_prompt_embeds,
        prompt_embeds_mask=prompt_embeds_mask,
        negative_prompt_embeds_mask=negative_prompt_embeds_mask,
        callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
        max_sequence_length=max_sequence_length,
    )

    self._guidance_scale = guidance_scale
    self._attention_kwargs = attention_kwargs
    self._current_timestep = None
    self._interrupt = False

    # 2. Define call parameters
    if prompt is not None and isinstance(prompt, str):
        batch_size = 1
    elif prompt is not None and isinstance(prompt, list):
        batch_size = len(prompt)
    else:
        batch_size = prompt_embeds.shape[0]

    has_neg_prompt = negative_prompt is not None or (
        negative_prompt_embeds is not None and negative_prompt_embeds_mask is not None
    )

    if true_cfg_scale > 1 and not has_neg_prompt:
        logger.warning(
            f"true_cfg_scale is passed as {true_cfg_scale}, but classifier-free guidance is not enabled since no negative_prompt is provided."
        )
    elif true_cfg_scale <= 1 and has_neg_prompt:
        logger.warning(
            " negative_prompt is passed but classifier-free guidance is not enabled since true_cfg_scale <= 1"
        )

    do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
    prompt_embeds, prompt_embeds_mask = self.encode_prompt(
        prompt=prompt,
        prompt_embeds=prompt_embeds,
        prompt_embeds_mask=prompt_embeds_mask,
        num_images_per_prompt=num_images_per_prompt,
        max_sequence_length=max_sequence_length,
    )
    if do_true_cfg:
        negative_prompt_embeds, negative_prompt_embeds_mask = self.encode_prompt(
            prompt=negative_prompt,
            prompt_embeds=negative_prompt_embeds,
            prompt_embeds_mask=negative_prompt_embeds_mask,
            num_images_per_prompt=num_images_per_prompt,
            max_sequence_length=max_sequence_length,
        )

    # 4. Prepare latent variables
    num_channels_latents = self.transformer.config.in_channels // 4
    latents = self.prepare_latents(
        batch_size * num_images_per_prompt,
        num_channels_latents,
        height,
        width,
        prompt_embeds.dtype,
        generator,
        latents,
    )
    img_shapes = [[(1, height // self.vae_scale_factor // 2, width // self.vae_scale_factor // 2)]] * batch_size

    # 5. Prepare timesteps
    sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else sigmas
    image_seq_len = latents.shape[1]
    mu = calculate_shift(
        image_seq_len,
        self.scheduler.config.get("base_image_seq_len", 256),
        self.scheduler.config.get("max_image_seq_len", 4096),
        self.scheduler.config.get("base_shift", 0.5),
        self.scheduler.config.get("max_shift", 1.15),
    )
    timesteps, num_inference_steps = retrieve_timesteps(
        self.scheduler,
        num_inference_steps,
        sigmas=sigmas,
        mu=mu,
    )
    num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
    self._num_timesteps = len(timesteps)

    # handle guidance
    if self.transformer.config.guidance_embeds and guidance_scale is None:
        raise ValueError("guidance_scale is required for guidance-distilled model.")
    elif self.transformer.config.guidance_embeds:
        guidance = mint.full([1], guidance_scale, dtype=ms.float32)
        guidance = guidance.expand((latents.shape[0],))
    elif not self.transformer.config.guidance_embeds and guidance_scale is not None:
        logger.warning(
            f"guidance_scale is passed as {guidance_scale}, but ignored since the model is not guidance-distilled."
        )
        guidance = None
    elif not self.transformer.config.guidance_embeds and guidance_scale is None:
        guidance = None

    if self.attention_kwargs is None:
        self._attention_kwargs = {}

    txt_seq_lens = prompt_embeds_mask.sum(dim=1).tolist() if prompt_embeds_mask is not None else None
    negative_txt_seq_lens = (
        negative_prompt_embeds_mask.sum(dim=1).tolist() if negative_prompt_embeds_mask is not None else None
    )

    # 6. Denoising loop
    self.scheduler.set_begin_index(0)
    with self.progress_bar(total=num_inference_steps) as progress_bar:
        for i, t in enumerate(timesteps):
            if self.interrupt:
                continue

            self._current_timestep = t
            # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
            timestep = t.expand((latents.shape[0],)).to(latents.dtype)
            noise_pred = self.transformer(
                hidden_states=latents,
                timestep=timestep / 1000,
                guidance=guidance,
                encoder_hidden_states_mask=prompt_embeds_mask,
                encoder_hidden_states=prompt_embeds,
                img_shapes=img_shapes,
                txt_seq_lens=txt_seq_lens,
                attention_kwargs=self.attention_kwargs,
                return_dict=False,
            )[0]

            if do_true_cfg:
                neg_noise_pred = self.transformer(
                    hidden_states=latents,
                    timestep=timestep / 1000,
                    guidance=guidance,
                    encoder_hidden_states_mask=negative_prompt_embeds_mask,
                    encoder_hidden_states=negative_prompt_embeds,
                    img_shapes=img_shapes,
                    txt_seq_lens=negative_txt_seq_lens,
                    attention_kwargs=self.attention_kwargs,
                    return_dict=False,
                )[0]
                comb_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)

                cond_norm = mint.norm(noise_pred, dim=-1, keepdim=True)
                noise_norm = mint.norm(comb_pred, dim=-1, keepdim=True)
                noise_pred = comb_pred * (cond_norm / noise_norm)

            # compute the previous noisy sample x_t -> x_t-1
            latents_dtype = latents.dtype
            latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]

            if latents.dtype != latents_dtype:
                latents = latents.to(latents_dtype)

            if callback_on_step_end is not None:
                callback_kwargs = {}
                for k in callback_on_step_end_tensor_inputs:
                    callback_kwargs[k] = locals()[k]
                callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)

                latents = callback_outputs.pop("latents", latents)
                prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)

            # call the callback, if provided
            if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                progress_bar.update()

    self._current_timestep = None
    if output_type == "latent":
        image = latents
    else:
        latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
        latents = latents.to(self.vae.dtype)
        latents_mean = (
            ms.tensor(self.vae.config.latents_mean).view(1, self.vae.config.z_dim, 1, 1, 1).to(latents.dtype)
        )
        latents_std = 1.0 / ms.tensor(self.vae.config.latents_std).view(1, self.vae.config.z_dim, 1, 1, 1).to(
            latents.dtype
        )
        latents = latents / latents_std + latents_mean
        image = self.vae.decode(latents, return_dict=False)[0][:, :, 0]
        image = self.image_processor.postprocess(image, output_type=output_type)

    if not return_dict:
        return (image,)

    return QwenImagePipelineOutput(images=image)

mindone.diffusers.QwenImagePipeline.disable_vae_slicing()

Disable sliced VAE decoding. If enable_vae_slicing was previously enabled, this method will go back to computing decoding in one step.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage.py

def disable_vae_slicing(self):
    r"""
    Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
    computing decoding in one step.
    """
    self.vae.disable_slicing()

mindone.diffusers.QwenImagePipeline.disable_vae_tiling()

Disable tiled VAE decoding. If enable_vae_tiling was previously enabled, this method will go back to computing decoding in one step.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage.py

def disable_vae_tiling(self):
    r"""
    Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
    computing decoding in one step.
    """
    self.vae.disable_tiling()

mindone.diffusers.QwenImagePipeline.enable_vae_slicing()

Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage.py

def enable_vae_slicing(self):
    r"""
    Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
    compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
    """
    self.vae.enable_slicing()

mindone.diffusers.QwenImagePipeline.enable_vae_tiling()

Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow processing larger images.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage.py

def enable_vae_tiling(self):
    r"""
    Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
    compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
    processing larger images.
    """
    self.vae.enable_tiling()

mindone.diffusers.QwenImagePipeline.encode_prompt(prompt, num_images_per_prompt=1, prompt_embeds=None, prompt_embeds_mask=None, max_sequence_length=1024)

PARAMETER	DESCRIPTION
prompt	prompt to be encoded TYPE: `str` or `List[str]`, *optional*
num_images_per_prompt	number of images that should be generated per prompt TYPE: `int` DEFAULT: 1
prompt_embeds	Pre-generated text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, text embeddings will be generated from prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage.py

def encode_prompt(
    self,
    prompt: Union[str, List[str]],
    num_images_per_prompt: int = 1,
    prompt_embeds: Optional[ms.Tensor] = None,
    prompt_embeds_mask: Optional[ms.Tensor] = None,
    max_sequence_length: int = 1024,
):
    r"""

    Args:
        prompt (`str` or `List[str]`, *optional*):
            prompt to be encoded
        num_images_per_prompt (`int`):
            number of images that should be generated per prompt
        prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
            provided, text embeddings will be generated from `prompt` input argument.
    """
    prompt = [prompt] if isinstance(prompt, str) else prompt
    batch_size = len(prompt) if prompt_embeds is None else prompt_embeds.shape[0]

    if prompt_embeds is None:
        prompt_embeds, prompt_embeds_mask = self._get_qwen_prompt_embeds(prompt)

    prompt_embeds = prompt_embeds[:, :max_sequence_length]
    prompt_embeds_mask = prompt_embeds_mask[:, :max_sequence_length]

    _, seq_len, _ = prompt_embeds.shape
    prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
    prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
    prompt_embeds_mask = prompt_embeds_mask.repeat(1, num_images_per_prompt, 1)
    prompt_embeds_mask = prompt_embeds_mask.view(batch_size * num_images_per_prompt, seq_len)

    return prompt_embeds, prompt_embeds_mask

mindone.diffusers.pipelines.qwenimage.pipeline_output.QwenImagePipelineOutput dataclass

Bases: BaseOutput

Output class for Stable Diffusion pipelines.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_output.py

@dataclass
class QwenImagePipelineOutput(BaseOutput):
    """
    Output class for Stable Diffusion pipelines.

    Args:
        images (`List[PIL.Image.Image]` or `np.ndarray`)
            List of denoised PIL images of length `batch_size` or numpy array of shape `(batch_size, height, width,
            num_channels)`. PIL images or numpy array present the denoised images of the diffusion pipeline.
    """

    images: Union[List[PIL.Image.Image], np.ndarray]

mindone.diffusers.QwenImageImg2ImgPipeline

Bases: DiffusionPipeline, QwenImageLoraLoaderMixin

The QwenImage pipeline for text-to-image generation.

PARAMETER	DESCRIPTION
transformer	Conditional Transformer (MMDiT) architecture to denoise the encoded image latents. TYPE: [`QwenImageTransformer2DModel`]
scheduler	A scheduler to be used in combination with transformer to denoise the encoded image latents. TYPE: [`FlowMatchEulerDiscreteScheduler`]
vae	Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. TYPE: [`AutoencoderKL`]
text_encoder	Qwen2.5-VL-7B-Instruct, specifically the Qwen2.5-VL-7B-Instruct variant. TYPE: [`Qwen2.5-VL-7B-Instruct`]
tokenizer	Tokenizer of class CLIPTokenizer. TYPE: `QwenTokenizer`

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py

class QwenImageImg2ImgPipeline(DiffusionPipeline, QwenImageLoraLoaderMixin):
    r"""
    The QwenImage pipeline for text-to-image generation.

    Args:
        transformer ([`QwenImageTransformer2DModel`]):
            Conditional Transformer (MMDiT) architecture to denoise the encoded image latents.
        scheduler ([`FlowMatchEulerDiscreteScheduler`]):
            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
        vae ([`AutoencoderKL`]):
            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
        text_encoder ([`Qwen2.5-VL-7B-Instruct`]):
            [Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct), specifically the
            [Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct) variant.
        tokenizer (`QwenTokenizer`):
            Tokenizer of class
            [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer).
    """

    model_cpu_offload_seq = "text_encoder->transformer->vae"
    _callback_tensor_inputs = ["latents", "prompt_embeds"]

    def __init__(
        self,
        scheduler: FlowMatchEulerDiscreteScheduler,
        vae: AutoencoderKLQwenImage,
        text_encoder: Qwen2_5_VLForConditionalGeneration,
        tokenizer: Qwen2Tokenizer,
        transformer: QwenImageTransformer2DModel,
    ):
        super().__init__()

        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            tokenizer=tokenizer,
            transformer=transformer,
            scheduler=scheduler,
        )
        self.vae_scale_factor = 2 ** len(self.vae.temperal_downsample) if getattr(self, "vae", None) else 8
        # QwenImage latents are turned into 2x2 patches and packed. This means the latent width and height has to be divisible
        # by the patch size. So the vae scale factor is multiplied by the patch size to account for this
        self.latent_channels = self.vae.config.z_dim if getattr(self, "vae", None) else 16
        self.image_processor = VaeImageProcessor(
            vae_scale_factor=self.vae_scale_factor * 2, vae_latent_channels=self.latent_channels
        )
        self.tokenizer_max_length = 1024
        self.prompt_template_encode = (
            "<|im_start|>system\nDescribe the image by detailing the color, shape, size, "
            "texture, quantity, text, spatial relationships of the objects and background:"
            "<|im_end|>\n<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
        )
        self.prompt_template_encode_start_idx = 34
        self.default_sample_size = 128

    # Copied from diffusers.pipelines.qwenimage.pipeline_qwenimage.QwenImagePipeline._extract_masked_hidden
    def _extract_masked_hidden(self, hidden_states: ms.Tensor, mask: ms.Tensor):
        bool_mask = mask.bool()
        valid_lengths = bool_mask.sum(dim=1)
        selected = hidden_states[bool_mask]
        split_result = mint.split(selected, valid_lengths.tolist(), dim=0)

        return split_result

    # Copied from diffusers.pipelines.qwenimage.pipeline_qwenimage.QwenImagePipeline._get_qwen_prompt_embeds
    def _get_qwen_prompt_embeds(
        self,
        prompt: Union[str, List[str]] = None,
        dtype: Optional[ms.Type] = None,
    ):
        dtype = dtype or self.text_encoder.dtype

        prompt = [prompt] if isinstance(prompt, str) else prompt

        template = self.prompt_template_encode
        drop_idx = self.prompt_template_encode_start_idx
        txt = [template.format(e) for e in prompt]
        txt_tokens = self.tokenizer(
            txt, max_length=self.tokenizer_max_length + drop_idx, padding=True, truncation=True, return_tensors="np"
        )
        encoder_hidden_states = self.text_encoder(
            input_ids=ms.tensor(txt_tokens.input_ids),
            attention_mask=ms.tensor(txt_tokens.attention_mask),
            output_hidden_states=True,
        )
        hidden_states = encoder_hidden_states.hidden_states[-1]
        split_hidden_states = self._extract_masked_hidden(hidden_states, ms.tensor(txt_tokens.attention_mask))
        split_hidden_states = [e[drop_idx:] for e in split_hidden_states]
        attn_mask_list = [mint.ones(e.shape[0], dtype=ms.int64) for e in split_hidden_states]
        max_seq_len = max([e.shape[0] for e in split_hidden_states])
        prompt_embeds = mint.stack(
            [mint.cat([u, u.new_zeros((max_seq_len - u.shape[0], u.shape[1]))]) for u in split_hidden_states]
        )
        encoder_attention_mask = mint.stack(
            [mint.cat([u, u.new_zeros((max_seq_len - u.shape[0]))]) for u in attn_mask_list]
        )

        prompt_embeds = prompt_embeds.to(dtype=dtype)

        return prompt_embeds, encoder_attention_mask

    def _encode_vae_image(self, image: ms.Tensor, generator: np.random.Generator):
        if isinstance(generator, list):
            image_latents = [
                retrieve_latents(self.vae, self.vae.encode(image[i : i + 1])[0]) for i in range(image.shape[0])
            ]
            image_latents = mint.cat(image_latents, dim=0)
        else:
            image_latents = retrieve_latents(self.vae, self.vae.encode(image)[0])

        latents_mean = (
            ms.tensor(self.vae.config.latents_mean).view(1, self.vae.config.z_dim, 1, 1, 1).to(image_latents.dtype)
        )
        latents_std = 1.0 / ms.tensor(self.vae.config.latents_std).view(1, self.vae.config.z_dim, 1, 1, 1).to(
            image_latents.dtype
        )

        image_latents = (image_latents - latents_mean) * latents_std

        return image_latents

    # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img.StableDiffusion3Img2ImgPipeline.get_timesteps
    def get_timesteps(self, num_inference_steps, strength):
        # get the original timestep using init_timestep
        init_timestep = min(num_inference_steps * strength, num_inference_steps)

        t_start = int(max(num_inference_steps - init_timestep, 0))
        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
        if hasattr(self.scheduler, "set_begin_index"):
            self.scheduler.set_begin_index(t_start * self.scheduler.order)

        return timesteps, num_inference_steps - t_start

    # Copied fromCopied from diffusers.pipelines.qwenimage.pipeline_qwenimage.QwenImagePipeline.encode_prompt
    def encode_prompt(
        self,
        prompt: Union[str, List[str]],
        num_images_per_prompt: int = 1,
        prompt_embeds: Optional[ms.Tensor] = None,
        prompt_embeds_mask: Optional[ms.Tensor] = None,
        max_sequence_length: int = 1024,
    ):
        r"""

        Args:
            prompt (`str` or `List[str]`, *optional*):
                prompt to be encoded
            num_images_per_prompt (`int`):
                number of images that should be generated per prompt
            prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
        """
        prompt = [prompt] if isinstance(prompt, str) else prompt
        batch_size = len(prompt) if prompt_embeds is None else prompt_embeds.shape[0]

        if prompt_embeds is None:
            prompt_embeds, prompt_embeds_mask = self._get_qwen_prompt_embeds(prompt)

        prompt_embeds = prompt_embeds[:, :max_sequence_length]
        prompt_embeds_mask = prompt_embeds_mask[:, :max_sequence_length]

        _, seq_len, _ = prompt_embeds.shape
        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
        prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
        prompt_embeds_mask = prompt_embeds_mask.repeat(1, num_images_per_prompt, 1)
        prompt_embeds_mask = prompt_embeds_mask.view(batch_size * num_images_per_prompt, seq_len)

        return prompt_embeds, prompt_embeds_mask

    def check_inputs(
        self,
        prompt,
        strength,
        height,
        width,
        negative_prompt=None,
        prompt_embeds=None,
        negative_prompt_embeds=None,
        prompt_embeds_mask=None,
        negative_prompt_embeds_mask=None,
        callback_on_step_end_tensor_inputs=None,
        max_sequence_length=None,
    ):
        if strength < 0 or strength > 1:
            raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")

        if height % (self.vae_scale_factor * 2) != 0 or width % (self.vae_scale_factor * 2) != 0:
            logger.warning(
                f"`height` and `width` have to be divisible by {self.vae_scale_factor * 2} but are {height} and {width}."
                " Dimensions will be resized accordingly"
            )

        if callback_on_step_end_tensor_inputs is not None and not all(
            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
        ):
            raise ValueError(
                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found"
                f" {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
            )

        if prompt is not None and prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt is None and prompt_embeds is None:
            raise ValueError(
                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
            )
        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")

        if negative_prompt is not None and negative_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
            )

        if prompt_embeds is not None and prompt_embeds_mask is None:
            raise ValueError(
                "If `prompt_embeds` are provided, `prompt_embeds_mask` also have to be passed. Make sure to generate `prompt_embeds_mask`"
                " from the same text encoder that was used to generate `prompt_embeds`."
            )

        if negative_prompt_embeds is not None and negative_prompt_embeds_mask is None:
            raise ValueError(
                "If `negative_prompt_embeds` are provided, `negative_prompt_embeds_mask` also have to be passed. Make sure to generate"
                " `negative_prompt_embeds_mask` from the same text encoder that was used to generate `negative_prompt_embeds`."
            )

        if max_sequence_length is not None and max_sequence_length > 1024:
            raise ValueError(f"`max_sequence_length` cannot be greater than 1024 but is {max_sequence_length}")

    @staticmethod
    # Copied from diffusers.pipelines.qwenimage.pipeline_qwenimage.QwenImagePipeline._pack_latents
    def _pack_latents(latents, batch_size, num_channels_latents, height, width):
        latents = latents.view(batch_size, num_channels_latents, height // 2, 2, width // 2, 2)
        latents = latents.permute(0, 2, 4, 1, 3, 5)
        latents = latents.reshape(batch_size, (height // 2) * (width // 2), num_channels_latents * 4)

        return latents

    @staticmethod
    # Copied from diffusers.pipelines.qwenimage.pipeline_qwenimage.QwenImagePipeline._unpack_latents
    def _unpack_latents(latents, height, width, vae_scale_factor):
        batch_size, num_patches, channels = latents.shape

        # VAE applies 8x compression on images but we must also account for packing which requires
        # latent height and width to be divisible by 2.
        height = 2 * (int(height) // (vae_scale_factor * 2))
        width = 2 * (int(width) // (vae_scale_factor * 2))

        latents = latents.view(batch_size, height // 2, width // 2, channels // 4, 2, 2)
        latents = latents.permute(0, 3, 1, 4, 2, 5)

        latents = latents.reshape(batch_size, channels // (2 * 2), 1, height, width)

        return latents

    def enable_vae_slicing(self):
        r"""
        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
        """
        self.vae.enable_slicing()

    def disable_vae_slicing(self):
        r"""
        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
        computing decoding in one step.
        """
        self.vae.disable_slicing()

    def enable_vae_tiling(self):
        r"""
        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
        processing larger images.
        """
        self.vae.enable_tiling()

    def disable_vae_tiling(self):
        r"""
        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
        computing decoding in one step.
        """
        self.vae.disable_tiling()

    def prepare_latents(
        self,
        image,
        timestep,
        batch_size,
        num_channels_latents,
        height,
        width,
        dtype,
        generator,
        latents=None,
    ):
        if isinstance(generator, list) and len(generator) != batch_size:
            raise ValueError(
                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
            )
        # VAE applies 8x compression on images but we must also account for packing which requires
        # latent height and width to be divisible by 2.
        height = 2 * (int(height) // (self.vae_scale_factor * 2))
        width = 2 * (int(width) // (self.vae_scale_factor * 2))

        shape = (batch_size, 1, num_channels_latents, height, width)

        # If image is [B,C,H,W] -> add T=1. If it's already [B,C,T,H,W], leave it.
        if image.dim() == 4:
            image = image.unsqueeze(2)
        elif image.dim() != 5:
            raise ValueError(f"Expected image dims 4 or 5, got {image.dim()}.")

        if latents is not None:
            return latents.to(dtype=dtype)

        image = image.to(dtype=dtype)
        if image.shape[1] != self.latent_channels:
            image_latents = self._encode_vae_image(image=image, generator=generator)  # [B,z,1,H',W']
        else:
            image_latents = image
        if batch_size > image_latents.shape[0] and batch_size % image_latents.shape[0] == 0:
            # expand init_latents for batch_size
            additional_image_per_prompt = batch_size // image_latents.shape[0]
            image_latents = mint.cat([image_latents] * additional_image_per_prompt, dim=0)
        elif batch_size > image_latents.shape[0] and batch_size % image_latents.shape[0] != 0:
            raise ValueError(
                f"Cannot duplicate `image` of batch size {image_latents.shape[0]} to {batch_size} text prompts."
            )
        else:
            image_latents = mint.cat([image_latents], dim=0)

        image_latents = image_latents.transpose(1, 2)  # [B,1,z,H',W']
        noise = randn_tensor(shape, generator=generator, dtype=dtype)
        latents = self.scheduler.scale_noise(image_latents, timestep, noise)
        latents = self._pack_latents(latents, batch_size, num_channels_latents, height, width)

        return latents

    @property
    def guidance_scale(self):
        return self._guidance_scale

    @property
    def attention_kwargs(self):
        return self._attention_kwargs

    @property
    def num_timesteps(self):
        return self._num_timesteps

    @property
    def current_timestep(self):
        return self._current_timestep

    @property
    def interrupt(self):
        return self._interrupt

    def __call__(
        self,
        prompt: Union[str, List[str]] = None,
        negative_prompt: Union[str, List[str]] = None,
        true_cfg_scale: float = 4.0,
        image: PipelineImageInput = None,
        height: Optional[int] = None,
        width: Optional[int] = None,
        strength: float = 0.6,
        num_inference_steps: int = 50,
        sigmas: Optional[List[float]] = None,
        guidance_scale: Optional[float] = None,
        num_images_per_prompt: int = 1,
        generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
        latents: Optional[ms.Tensor] = None,
        prompt_embeds: Optional[ms.Tensor] = None,
        prompt_embeds_mask: Optional[ms.Tensor] = None,
        negative_prompt_embeds: Optional[ms.Tensor] = None,
        negative_prompt_embeds_mask: Optional[ms.Tensor] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = True,
        attention_kwargs: Optional[Dict[str, Any]] = None,
        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
        max_sequence_length: int = 512,
    ):
        r"""
        Function invoked when calling the pipeline for generation.

        Args:
            prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
                instead.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation. If not defined, one has to pass
                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
                not greater than `1`).
            image (`ms.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[ms.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`):
                `Image`, numpy array or tensor representing an image batch to be used as the starting point. For both
                numpy array and mindspore tensor, the expected value range is between `[0, 1]` If it's a tensor or a list
                or tensors, the expected shape should be `(B, C, H, W)` or `(C, H, W)`. If it is a numpy array or a
                list of arrays, the expected shape should be `(B, H, W, C)` or `(H, W, C)` It can also accept image
                latents as `image`, but if passing latents directly it is not encoded again.
            true_cfg_scale (`float`, *optional*, defaults to 1.0):
                Guidance scale as defined in [Classifier-Free Diffusion
                Guidance](https://huggingface.co/papers/2207.12598). `true_cfg_scale` is defined as `w` of equation 2.
                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Classifier-free guidance is enabled by
                setting `true_cfg_scale > 1` and a provided `negative_prompt`. Higher guidance scale encourages to
                generate images that are closely linked to the text `prompt`, usually at the expense of lower image
                quality.
            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                The height in pixels of the generated image. This is set to 1024 by default for the best results.
            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                The width in pixels of the generated image. This is set to 1024 by default for the best results.
            strength (`float`, *optional*, defaults to 1.0):
                Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a
                starting point and more noise is added the higher the `strength`. The number of denoising steps depends
                on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising
                process runs for the full number of iterations specified in `num_inference_steps`. A value of 1
                essentially ignores `image`.
            num_inference_steps (`int`, *optional*, defaults to 50):
                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                expense of slower inference.
            sigmas (`List[float]`, *optional*):
                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                will be used.
            guidance_scale (`float`, *optional*, defaults to None):
                A guidance scale value for guidance distilled models. Unlike the traditional classifier-free guidance
                where the guidance scale is applied during inference through noise prediction rescaling, guidance
                distilled models take the guidance scale directly as an input parameter during forward pass. Guidance
                scale is enabled by setting `guidance_scale > 1`. Higher guidance scale encourages to generate images
                that are closely linked to the text `prompt`, usually at the expense of lower image quality. This
                parameter in the pipeline is there to support future guidance-distilled models when they come up. It is
                ignored when not using guidance distilled models. To enable traditional classifier-free guidance,
                please pass `true_cfg_scale > 1.0` and `negative_prompt` (even an empty negative prompt like " " should
                enable classifier-free guidance computations).
            num_images_per_prompt (`int`, *optional*, defaults to 1):
                The number of images to generate per prompt.
            generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
                One or a list of [numpy generator(s)](https://numpy.org/doc/stable/reference/random/generator.html)
                to make generation deterministic.
            latents (`ms.Tensor`, *optional*):
                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor will be generated by sampling using the supplied random `generator`.
            prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
            negative_prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                argument.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generate image. Choose between
                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.qwenimage.QwenImagePipelineOutput`] instead of a plain tuple.
            attention_kwargs (`dict`, *optional*):
                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
                `self.processor` in
                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
            callback_on_step_end (`Callable`, *optional*):
                A function that calls at the end of each denoising steps during the inference. The function is called
                with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
                callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
                `callback_on_step_end_tensor_inputs`.
            callback_on_step_end_tensor_inputs (`List`, *optional*):
                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
                `._callback_tensor_inputs` attribute of your pipeline class.
            max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.

        Examples:

        Returns:
            [`~pipelines.qwenimage.QwenImagePipelineOutput`] or `tuple`:
            [`~pipelines.qwenimage.QwenImagePipelineOutput`] if `return_dict` is True, otherwise a `tuple`. When
            returning a tuple, the first element is a list with the generated images.
        """

        height = height or self.default_sample_size * self.vae_scale_factor
        width = width or self.default_sample_size * self.vae_scale_factor

        # 1. Check inputs. Raise error if not correct
        self.check_inputs(
            prompt,
            strength,
            height,
            width,
            negative_prompt=negative_prompt,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            prompt_embeds_mask=prompt_embeds_mask,
            negative_prompt_embeds_mask=negative_prompt_embeds_mask,
            callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
            max_sequence_length=max_sequence_length,
        )

        self._guidance_scale = guidance_scale
        self._attention_kwargs = attention_kwargs
        self._current_timestep = None
        self._interrupt = False

        # 2. Preprocess image
        init_image = self.image_processor.preprocess(image, height=height, width=width)
        init_image = init_image.to(dtype=ms.float32)

        # 3. Define call parameters
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        has_neg_prompt = negative_prompt is not None or (
            negative_prompt_embeds is not None and negative_prompt_embeds_mask is not None
        )

        if true_cfg_scale > 1 and not has_neg_prompt:
            logger.warning(
                f"true_cfg_scale is passed as {true_cfg_scale}, but classifier-free guidance is not enabled since no negative_prompt is provided."
            )
        elif true_cfg_scale <= 1 and has_neg_prompt:
            logger.warning(
                " negative_prompt is passed but classifier-free guidance is not enabled since true_cfg_scale <= 1"
            )

        do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
        prompt_embeds, prompt_embeds_mask = self.encode_prompt(
            prompt=prompt,
            prompt_embeds=prompt_embeds,
            prompt_embeds_mask=prompt_embeds_mask,
            num_images_per_prompt=num_images_per_prompt,
            max_sequence_length=max_sequence_length,
        )
        if do_true_cfg:
            negative_prompt_embeds, negative_prompt_embeds_mask = self.encode_prompt(
                prompt=negative_prompt,
                prompt_embeds=negative_prompt_embeds,
                prompt_embeds_mask=negative_prompt_embeds_mask,
                num_images_per_prompt=num_images_per_prompt,
                max_sequence_length=max_sequence_length,
            )

        # 4. Prepare timesteps
        sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else sigmas
        image_seq_len = (int(height) // self.vae_scale_factor // 2) * (int(width) // self.vae_scale_factor // 2)
        mu = calculate_shift(
            image_seq_len,
            self.scheduler.config.get("base_image_seq_len", 256),
            self.scheduler.config.get("max_image_seq_len", 4096),
            self.scheduler.config.get("base_shift", 0.5),
            self.scheduler.config.get("max_shift", 1.15),
        )
        timesteps, num_inference_steps = retrieve_timesteps(
            self.scheduler,
            num_inference_steps,
            sigmas=sigmas,
            mu=mu,
        )
        timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength)
        if num_inference_steps < 1:
            raise ValueError(
                f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline"
                f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline."
            )
        latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)

        # 5. Prepare latent variables
        num_channels_latents = self.transformer.config.in_channels // 4
        latents = self.prepare_latents(
            init_image,
            latent_timestep,
            batch_size * num_images_per_prompt,
            num_channels_latents,
            height,
            width,
            prompt_embeds.dtype,
            generator,
            latents,
        )
        img_shapes = [[(1, height // self.vae_scale_factor // 2, width // self.vae_scale_factor // 2)]] * batch_size

        num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
        self._num_timesteps = len(timesteps)

        # handle guidance
        if self.transformer.config.guidance_embeds and guidance_scale is None:
            raise ValueError("guidance_scale is required for guidance-distilled model.")
        elif self.transformer.config.guidance_embeds:
            guidance = mint.full([1], guidance_scale, dtype=ms.float32)
            guidance = guidance.expand((latents.shape[0],))
        elif not self.transformer.config.guidance_embeds and guidance_scale is not None:
            logger.warning(
                f"guidance_scale is passed as {guidance_scale}, but ignored since the model is not guidance-distilled."
            )
            guidance = None
        elif not self.transformer.config.guidance_embeds and guidance_scale is None:
            guidance = None

        if self.attention_kwargs is None:
            self._attention_kwargs = {}

        txt_seq_lens = prompt_embeds_mask.sum(dim=1).tolist() if prompt_embeds_mask is not None else None
        negative_txt_seq_lens = (
            negative_prompt_embeds_mask.sum(dim=1).tolist() if negative_prompt_embeds_mask is not None else None
        )

        # 6. Denoising loop
        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(timesteps):
                if self.interrupt:
                    continue

                self._current_timestep = t
                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
                timestep = t.expand((latents.shape[0],)).to(latents.dtype)
                noise_pred = self.transformer(
                    hidden_states=latents,
                    timestep=timestep / 1000,
                    guidance=guidance,
                    encoder_hidden_states_mask=prompt_embeds_mask,
                    encoder_hidden_states=prompt_embeds,
                    img_shapes=img_shapes,
                    txt_seq_lens=txt_seq_lens,
                    attention_kwargs=self.attention_kwargs,
                    return_dict=False,
                )[0]

                if do_true_cfg:
                    neg_noise_pred = self.transformer(
                        hidden_states=latents,
                        timestep=timestep / 1000,
                        guidance=guidance,
                        encoder_hidden_states_mask=negative_prompt_embeds_mask,
                        encoder_hidden_states=negative_prompt_embeds,
                        img_shapes=img_shapes,
                        txt_seq_lens=negative_txt_seq_lens,
                        attention_kwargs=self.attention_kwargs,
                        return_dict=False,
                    )[0]
                    comb_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)

                    cond_norm = mint.norm(noise_pred, dim=-1, keepdim=True)
                    noise_norm = mint.norm(comb_pred, dim=-1, keepdim=True)
                    noise_pred = comb_pred * (cond_norm / noise_norm)

                # compute the previous noisy sample x_t -> x_t-1
                latents_dtype = latents.dtype
                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]

                if latents.dtype != latents_dtype:
                    latents = latents.to(latents_dtype)

                if callback_on_step_end is not None:
                    callback_kwargs = {}
                    for k in callback_on_step_end_tensor_inputs:
                        callback_kwargs[k] = locals()[k]
                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)

                    latents = callback_outputs.pop("latents", latents)
                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)

                # call the callback, if provided
                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                    progress_bar.update()

        self._current_timestep = None
        if output_type == "latent":
            image = latents
        else:
            latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
            latents = latents.to(self.vae.dtype)
            latents_mean = (
                ms.tensor(self.vae.config.latents_mean).view(1, self.vae.config.z_dim, 1, 1, 1).to(latents.dtype)
            )
            latents_std = 1.0 / ms.tensor(self.vae.config.latents_std).view(1, self.vae.config.z_dim, 1, 1, 1).to(
                latents.dtype
            )

            latents = latents / latents_std + latents_mean
            image = self.vae.decode(latents, return_dict=False)[0][:, :, 0]
            image = self.image_processor.postprocess(image, output_type=output_type)

        if not return_dict:
            return (image,)

        return QwenImagePipelineOutput(images=image)

mindone.diffusers.QwenImageImg2ImgPipeline.__call__(prompt=None, negative_prompt=None, true_cfg_scale=4.0, image=None, height=None, width=None, strength=0.6, num_inference_steps=50, sigmas=None, guidance_scale=None, num_images_per_prompt=1, generator=None, latents=None, prompt_embeds=None, prompt_embeds_mask=None, negative_prompt_embeds=None, negative_prompt_embeds_mask=None, output_type='pil', return_dict=True, attention_kwargs=None, callback_on_step_end=None, callback_on_step_end_tensor_inputs=['latents'], max_sequence_length=512)

Function invoked when calling the pipeline for generation.

PARAMETER	DESCRIPTION
prompt	The prompt or prompts to guide the image generation. If not defined, one has to pass prompt_embeds. instead. TYPE: `str` or `List[str]`, *optional* DEFAULT: None
negative_prompt	The prompt or prompts not to guide the image generation. If not defined, one has to pass negative_prompt_embeds instead. Ignored when not using guidance (i.e., ignored if true_cfg_scale is not greater than 1). TYPE: `str` or `List[str]`, *optional* DEFAULT: None
image	Image, numpy array or tensor representing an image batch to be used as the starting point. For both numpy array and mindspore tensor, the expected value range is between [0, 1] If it's a tensor or a list or tensors, the expected shape should be (B, C, H, W) or (C, H, W). If it is a numpy array or a list of arrays, the expected shape should be (B, H, W, C) or (H, W, C) It can also accept image latents as image, but if passing latents directly it is not encoded again. TYPE: `ms.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[ms.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]` DEFAULT: None
true_cfg_scale	Guidance scale as defined in Classifier-Free Diffusion Guidance. true_cfg_scale is defined as w of equation 2. of Imagen Paper. Classifier-free guidance is enabled by setting true_cfg_scale > 1 and a provided negative_prompt. Higher guidance scale encourages to generate images that are closely linked to the text prompt, usually at the expense of lower image quality. TYPE: `float`, *optional*, defaults to 1.0 DEFAULT: 4.0
height	The height in pixels of the generated image. This is set to 1024 by default for the best results. TYPE: `int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor DEFAULT: None
width	The width in pixels of the generated image. This is set to 1024 by default for the best results. TYPE: `int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor DEFAULT: None
strength	Indicates extent to transform the reference image. Must be between 0 and 1. image is used as a starting point and more noise is added the higher the strength. The number of denoising steps depends on the amount of noise initially added. When strength is 1, added noise is maximum and the denoising process runs for the full number of iterations specified in num_inference_steps. A value of 1 essentially ignores image. TYPE: `float`, *optional*, defaults to 1.0 DEFAULT: 0.6
num_inference_steps	The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. TYPE: `int`, *optional*, defaults to 50 DEFAULT: 50
sigmas	Custom sigmas to use for the denoising process with schedulers which support a sigmas argument in their set_timesteps method. If not defined, the default behavior when num_inference_steps is passed will be used. TYPE: `List[float]`, *optional* DEFAULT: None
guidance_scale	A guidance scale value for guidance distilled models. Unlike the traditional classifier-free guidance where the guidance scale is applied during inference through noise prediction rescaling, guidance distilled models take the guidance scale directly as an input parameter during forward pass. Guidance scale is enabled by setting guidance_scale > 1. Higher guidance scale encourages to generate images that are closely linked to the text prompt, usually at the expense of lower image quality. This parameter in the pipeline is there to support future guidance-distilled models when they come up. It is ignored when not using guidance distilled models. To enable traditional classifier-free guidance, please pass true_cfg_scale > 1.0 and negative_prompt (even an empty negative prompt like " " should enable classifier-free guidance computations). TYPE: `float`, *optional*, defaults to None DEFAULT: None
num_images_per_prompt	The number of images to generate per prompt. TYPE: `int`, *optional*, defaults to 1 DEFAULT: 1
generator	One or a list of numpy generator(s) to make generation deterministic. TYPE: `np.random.Generator` or `List[np.random.Generator]`, *optional* DEFAULT: None
latents	Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor will be generated by sampling using the supplied random generator. TYPE: `ms.Tensor`, *optional* DEFAULT: None
prompt_embeds	Pre-generated text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, text embeddings will be generated from prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
negative_prompt_embeds	Pre-generated negative text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, negative_prompt_embeds will be generated from negative_prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
output_type	The output format of the generate image. Choose between PIL: PIL.Image.Image or np.array. TYPE: `str`, *optional*, defaults to `"pil"` DEFAULT: 'pil'
return_dict	Whether or not to return a [~pipelines.qwenimage.QwenImagePipelineOutput] instead of a plain tuple. TYPE: `bool`, *optional*, defaults to `True` DEFAULT: True
attention_kwargs	A kwargs dictionary that if specified is passed along to the AttentionProcessor as defined under self.processor in diffusers.models.attention_processor. TYPE: `dict`, *optional* DEFAULT: None
callback_on_step_end	A function that calls at the end of each denoising steps during the inference. The function is called with the following arguments: callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict). callback_kwargs will include a list of all tensors as specified by callback_on_step_end_tensor_inputs. TYPE: `Callable`, *optional* DEFAULT: None
callback_on_step_end_tensor_inputs	The list of tensor inputs for the callback_on_step_end function. The tensors specified in the list will be passed as callback_kwargs argument. You will only be able to include variables listed in the ._callback_tensor_inputs attribute of your pipeline class. TYPE: `List`, *optional* DEFAULT: ['latents']
max_sequence_length	Maximum sequence length to use with the prompt. TYPE: `int` defaults to 512 DEFAULT: 512

RETURNS	DESCRIPTION
	[~pipelines.qwenimage.QwenImagePipelineOutput] or tuple:
	[~pipelines.qwenimage.QwenImagePipelineOutput] if return_dict is True, otherwise a tuple. When
	returning a tuple, the first element is a list with the generated images.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py

def __call__(
    self,
    prompt: Union[str, List[str]] = None,
    negative_prompt: Union[str, List[str]] = None,
    true_cfg_scale: float = 4.0,
    image: PipelineImageInput = None,
    height: Optional[int] = None,
    width: Optional[int] = None,
    strength: float = 0.6,
    num_inference_steps: int = 50,
    sigmas: Optional[List[float]] = None,
    guidance_scale: Optional[float] = None,
    num_images_per_prompt: int = 1,
    generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
    latents: Optional[ms.Tensor] = None,
    prompt_embeds: Optional[ms.Tensor] = None,
    prompt_embeds_mask: Optional[ms.Tensor] = None,
    negative_prompt_embeds: Optional[ms.Tensor] = None,
    negative_prompt_embeds_mask: Optional[ms.Tensor] = None,
    output_type: Optional[str] = "pil",
    return_dict: bool = True,
    attention_kwargs: Optional[Dict[str, Any]] = None,
    callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
    callback_on_step_end_tensor_inputs: List[str] = ["latents"],
    max_sequence_length: int = 512,
):
    r"""
    Function invoked when calling the pipeline for generation.

    Args:
        prompt (`str` or `List[str]`, *optional*):
            The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
            instead.
        negative_prompt (`str` or `List[str]`, *optional*):
            The prompt or prompts not to guide the image generation. If not defined, one has to pass
            `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
            not greater than `1`).
        image (`ms.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[ms.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`):
            `Image`, numpy array or tensor representing an image batch to be used as the starting point. For both
            numpy array and mindspore tensor, the expected value range is between `[0, 1]` If it's a tensor or a list
            or tensors, the expected shape should be `(B, C, H, W)` or `(C, H, W)`. If it is a numpy array or a
            list of arrays, the expected shape should be `(B, H, W, C)` or `(H, W, C)` It can also accept image
            latents as `image`, but if passing latents directly it is not encoded again.
        true_cfg_scale (`float`, *optional*, defaults to 1.0):
            Guidance scale as defined in [Classifier-Free Diffusion
            Guidance](https://huggingface.co/papers/2207.12598). `true_cfg_scale` is defined as `w` of equation 2.
            of [Imagen Paper](https://huggingface.co/papers/2205.11487). Classifier-free guidance is enabled by
            setting `true_cfg_scale > 1` and a provided `negative_prompt`. Higher guidance scale encourages to
            generate images that are closely linked to the text `prompt`, usually at the expense of lower image
            quality.
        height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
            The height in pixels of the generated image. This is set to 1024 by default for the best results.
        width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
            The width in pixels of the generated image. This is set to 1024 by default for the best results.
        strength (`float`, *optional*, defaults to 1.0):
            Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a
            starting point and more noise is added the higher the `strength`. The number of denoising steps depends
            on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising
            process runs for the full number of iterations specified in `num_inference_steps`. A value of 1
            essentially ignores `image`.
        num_inference_steps (`int`, *optional*, defaults to 50):
            The number of denoising steps. More denoising steps usually lead to a higher quality image at the
            expense of slower inference.
        sigmas (`List[float]`, *optional*):
            Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
            their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
            will be used.
        guidance_scale (`float`, *optional*, defaults to None):
            A guidance scale value for guidance distilled models. Unlike the traditional classifier-free guidance
            where the guidance scale is applied during inference through noise prediction rescaling, guidance
            distilled models take the guidance scale directly as an input parameter during forward pass. Guidance
            scale is enabled by setting `guidance_scale > 1`. Higher guidance scale encourages to generate images
            that are closely linked to the text `prompt`, usually at the expense of lower image quality. This
            parameter in the pipeline is there to support future guidance-distilled models when they come up. It is
            ignored when not using guidance distilled models. To enable traditional classifier-free guidance,
            please pass `true_cfg_scale > 1.0` and `negative_prompt` (even an empty negative prompt like " " should
            enable classifier-free guidance computations).
        num_images_per_prompt (`int`, *optional*, defaults to 1):
            The number of images to generate per prompt.
        generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
            One or a list of [numpy generator(s)](https://numpy.org/doc/stable/reference/random/generator.html)
            to make generation deterministic.
        latents (`ms.Tensor`, *optional*):
            Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
            generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
            tensor will be generated by sampling using the supplied random `generator`.
        prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
            provided, text embeddings will be generated from `prompt` input argument.
        negative_prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
            weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
            argument.
        output_type (`str`, *optional*, defaults to `"pil"`):
            The output format of the generate image. Choose between
            [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
        return_dict (`bool`, *optional*, defaults to `True`):
            Whether or not to return a [`~pipelines.qwenimage.QwenImagePipelineOutput`] instead of a plain tuple.
        attention_kwargs (`dict`, *optional*):
            A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
            `self.processor` in
            [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
        callback_on_step_end (`Callable`, *optional*):
            A function that calls at the end of each denoising steps during the inference. The function is called
            with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
            callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
            `callback_on_step_end_tensor_inputs`.
        callback_on_step_end_tensor_inputs (`List`, *optional*):
            The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
            will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
            `._callback_tensor_inputs` attribute of your pipeline class.
        max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.

    Examples:

    Returns:
        [`~pipelines.qwenimage.QwenImagePipelineOutput`] or `tuple`:
        [`~pipelines.qwenimage.QwenImagePipelineOutput`] if `return_dict` is True, otherwise a `tuple`. When
        returning a tuple, the first element is a list with the generated images.
    """

    height = height or self.default_sample_size * self.vae_scale_factor
    width = width or self.default_sample_size * self.vae_scale_factor

    # 1. Check inputs. Raise error if not correct
    self.check_inputs(
        prompt,
        strength,
        height,
        width,
        negative_prompt=negative_prompt,
        prompt_embeds=prompt_embeds,
        negative_prompt_embeds=negative_prompt_embeds,
        prompt_embeds_mask=prompt_embeds_mask,
        negative_prompt_embeds_mask=negative_prompt_embeds_mask,
        callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
        max_sequence_length=max_sequence_length,
    )

    self._guidance_scale = guidance_scale
    self._attention_kwargs = attention_kwargs
    self._current_timestep = None
    self._interrupt = False

    # 2. Preprocess image
    init_image = self.image_processor.preprocess(image, height=height, width=width)
    init_image = init_image.to(dtype=ms.float32)

    # 3. Define call parameters
    if prompt is not None and isinstance(prompt, str):
        batch_size = 1
    elif prompt is not None and isinstance(prompt, list):
        batch_size = len(prompt)
    else:
        batch_size = prompt_embeds.shape[0]

    has_neg_prompt = negative_prompt is not None or (
        negative_prompt_embeds is not None and negative_prompt_embeds_mask is not None
    )

    if true_cfg_scale > 1 and not has_neg_prompt:
        logger.warning(
            f"true_cfg_scale is passed as {true_cfg_scale}, but classifier-free guidance is not enabled since no negative_prompt is provided."
        )
    elif true_cfg_scale <= 1 and has_neg_prompt:
        logger.warning(
            " negative_prompt is passed but classifier-free guidance is not enabled since true_cfg_scale <= 1"
        )

    do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
    prompt_embeds, prompt_embeds_mask = self.encode_prompt(
        prompt=prompt,
        prompt_embeds=prompt_embeds,
        prompt_embeds_mask=prompt_embeds_mask,
        num_images_per_prompt=num_images_per_prompt,
        max_sequence_length=max_sequence_length,
    )
    if do_true_cfg:
        negative_prompt_embeds, negative_prompt_embeds_mask = self.encode_prompt(
            prompt=negative_prompt,
            prompt_embeds=negative_prompt_embeds,
            prompt_embeds_mask=negative_prompt_embeds_mask,
            num_images_per_prompt=num_images_per_prompt,
            max_sequence_length=max_sequence_length,
        )

    # 4. Prepare timesteps
    sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else sigmas
    image_seq_len = (int(height) // self.vae_scale_factor // 2) * (int(width) // self.vae_scale_factor // 2)
    mu = calculate_shift(
        image_seq_len,
        self.scheduler.config.get("base_image_seq_len", 256),
        self.scheduler.config.get("max_image_seq_len", 4096),
        self.scheduler.config.get("base_shift", 0.5),
        self.scheduler.config.get("max_shift", 1.15),
    )
    timesteps, num_inference_steps = retrieve_timesteps(
        self.scheduler,
        num_inference_steps,
        sigmas=sigmas,
        mu=mu,
    )
    timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength)
    if num_inference_steps < 1:
        raise ValueError(
            f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline"
            f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline."
        )
    latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)

    # 5. Prepare latent variables
    num_channels_latents = self.transformer.config.in_channels // 4
    latents = self.prepare_latents(
        init_image,
        latent_timestep,
        batch_size * num_images_per_prompt,
        num_channels_latents,
        height,
        width,
        prompt_embeds.dtype,
        generator,
        latents,
    )
    img_shapes = [[(1, height // self.vae_scale_factor // 2, width // self.vae_scale_factor // 2)]] * batch_size

    num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
    self._num_timesteps = len(timesteps)

    # handle guidance
    if self.transformer.config.guidance_embeds and guidance_scale is None:
        raise ValueError("guidance_scale is required for guidance-distilled model.")
    elif self.transformer.config.guidance_embeds:
        guidance = mint.full([1], guidance_scale, dtype=ms.float32)
        guidance = guidance.expand((latents.shape[0],))
    elif not self.transformer.config.guidance_embeds and guidance_scale is not None:
        logger.warning(
            f"guidance_scale is passed as {guidance_scale}, but ignored since the model is not guidance-distilled."
        )
        guidance = None
    elif not self.transformer.config.guidance_embeds and guidance_scale is None:
        guidance = None

    if self.attention_kwargs is None:
        self._attention_kwargs = {}

    txt_seq_lens = prompt_embeds_mask.sum(dim=1).tolist() if prompt_embeds_mask is not None else None
    negative_txt_seq_lens = (
        negative_prompt_embeds_mask.sum(dim=1).tolist() if negative_prompt_embeds_mask is not None else None
    )

    # 6. Denoising loop
    with self.progress_bar(total=num_inference_steps) as progress_bar:
        for i, t in enumerate(timesteps):
            if self.interrupt:
                continue

            self._current_timestep = t
            # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
            timestep = t.expand((latents.shape[0],)).to(latents.dtype)
            noise_pred = self.transformer(
                hidden_states=latents,
                timestep=timestep / 1000,
                guidance=guidance,
                encoder_hidden_states_mask=prompt_embeds_mask,
                encoder_hidden_states=prompt_embeds,
                img_shapes=img_shapes,
                txt_seq_lens=txt_seq_lens,
                attention_kwargs=self.attention_kwargs,
                return_dict=False,
            )[0]

            if do_true_cfg:
                neg_noise_pred = self.transformer(
                    hidden_states=latents,
                    timestep=timestep / 1000,
                    guidance=guidance,
                    encoder_hidden_states_mask=negative_prompt_embeds_mask,
                    encoder_hidden_states=negative_prompt_embeds,
                    img_shapes=img_shapes,
                    txt_seq_lens=negative_txt_seq_lens,
                    attention_kwargs=self.attention_kwargs,
                    return_dict=False,
                )[0]
                comb_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)

                cond_norm = mint.norm(noise_pred, dim=-1, keepdim=True)
                noise_norm = mint.norm(comb_pred, dim=-1, keepdim=True)
                noise_pred = comb_pred * (cond_norm / noise_norm)

            # compute the previous noisy sample x_t -> x_t-1
            latents_dtype = latents.dtype
            latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]

            if latents.dtype != latents_dtype:
                latents = latents.to(latents_dtype)

            if callback_on_step_end is not None:
                callback_kwargs = {}
                for k in callback_on_step_end_tensor_inputs:
                    callback_kwargs[k] = locals()[k]
                callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)

                latents = callback_outputs.pop("latents", latents)
                prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)

            # call the callback, if provided
            if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                progress_bar.update()

    self._current_timestep = None
    if output_type == "latent":
        image = latents
    else:
        latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
        latents = latents.to(self.vae.dtype)
        latents_mean = (
            ms.tensor(self.vae.config.latents_mean).view(1, self.vae.config.z_dim, 1, 1, 1).to(latents.dtype)
        )
        latents_std = 1.0 / ms.tensor(self.vae.config.latents_std).view(1, self.vae.config.z_dim, 1, 1, 1).to(
            latents.dtype
        )

        latents = latents / latents_std + latents_mean
        image = self.vae.decode(latents, return_dict=False)[0][:, :, 0]
        image = self.image_processor.postprocess(image, output_type=output_type)

    if not return_dict:
        return (image,)

    return QwenImagePipelineOutput(images=image)

mindone.diffusers.QwenImageImg2ImgPipeline.disable_vae_slicing()

Disable sliced VAE decoding. If enable_vae_slicing was previously enabled, this method will go back to computing decoding in one step.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py

def disable_vae_slicing(self):
    r"""
    Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
    computing decoding in one step.
    """
    self.vae.disable_slicing()

mindone.diffusers.QwenImageImg2ImgPipeline.disable_vae_tiling()

Disable tiled VAE decoding. If enable_vae_tiling was previously enabled, this method will go back to computing decoding in one step.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py

def disable_vae_tiling(self):
    r"""
    Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
    computing decoding in one step.
    """
    self.vae.disable_tiling()

mindone.diffusers.QwenImageImg2ImgPipeline.enable_vae_slicing()

Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py

def enable_vae_slicing(self):
    r"""
    Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
    compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
    """
    self.vae.enable_slicing()

mindone.diffusers.QwenImageImg2ImgPipeline.enable_vae_tiling()

Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow processing larger images.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py

def enable_vae_tiling(self):
    r"""
    Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
    compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
    processing larger images.
    """
    self.vae.enable_tiling()

mindone.diffusers.QwenImageImg2ImgPipeline.encode_prompt(prompt, num_images_per_prompt=1, prompt_embeds=None, prompt_embeds_mask=None, max_sequence_length=1024)

PARAMETER	DESCRIPTION
prompt	prompt to be encoded TYPE: `str` or `List[str]`, *optional*
num_images_per_prompt	number of images that should be generated per prompt TYPE: `int` DEFAULT: 1
prompt_embeds	Pre-generated text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, text embeddings will be generated from prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_img2img.py

def encode_prompt(
    self,
    prompt: Union[str, List[str]],
    num_images_per_prompt: int = 1,
    prompt_embeds: Optional[ms.Tensor] = None,
    prompt_embeds_mask: Optional[ms.Tensor] = None,
    max_sequence_length: int = 1024,
):
    r"""

    Args:
        prompt (`str` or `List[str]`, *optional*):
            prompt to be encoded
        num_images_per_prompt (`int`):
            number of images that should be generated per prompt
        prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
            provided, text embeddings will be generated from `prompt` input argument.
    """
    prompt = [prompt] if isinstance(prompt, str) else prompt
    batch_size = len(prompt) if prompt_embeds is None else prompt_embeds.shape[0]

    if prompt_embeds is None:
        prompt_embeds, prompt_embeds_mask = self._get_qwen_prompt_embeds(prompt)

    prompt_embeds = prompt_embeds[:, :max_sequence_length]
    prompt_embeds_mask = prompt_embeds_mask[:, :max_sequence_length]

    _, seq_len, _ = prompt_embeds.shape
    prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
    prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
    prompt_embeds_mask = prompt_embeds_mask.repeat(1, num_images_per_prompt, 1)
    prompt_embeds_mask = prompt_embeds_mask.view(batch_size * num_images_per_prompt, seq_len)

    return prompt_embeds, prompt_embeds_mask

mindone.diffusers.QwenImageInpaintPipeline

Bases: DiffusionPipeline, QwenImageLoraLoaderMixin

The QwenImage pipeline for text-to-image generation.

PARAMETER	DESCRIPTION
transformer	Conditional Transformer (MMDiT) architecture to denoise the encoded image latents. TYPE: [`QwenImageTransformer2DModel`]
scheduler	A scheduler to be used in combination with transformer to denoise the encoded image latents. TYPE: [`FlowMatchEulerDiscreteScheduler`]
vae	Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. TYPE: [`AutoencoderKL`]
text_encoder	Qwen2.5-VL-7B-Instruct, specifically the Qwen2.5-VL-7B-Instruct variant. TYPE: [`Qwen2.5-VL-7B-Instruct`]
tokenizer	Tokenizer of class CLIPTokenizer. TYPE: `QwenTokenizer`

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py

class QwenImageInpaintPipeline(DiffusionPipeline, QwenImageLoraLoaderMixin):
    r"""
    The QwenImage pipeline for text-to-image generation.

    Args:
        transformer ([`QwenImageTransformer2DModel`]):
            Conditional Transformer (MMDiT) architecture to denoise the encoded image latents.
        scheduler ([`FlowMatchEulerDiscreteScheduler`]):
            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
        vae ([`AutoencoderKL`]):
            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
        text_encoder ([`Qwen2.5-VL-7B-Instruct`]):
            [Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct), specifically the
            [Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct) variant.
        tokenizer (`QwenTokenizer`):
            Tokenizer of class
            [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer).
    """

    model_cpu_offload_seq = "text_encoder->transformer->vae"
    _callback_tensor_inputs = ["latents", "prompt_embeds"]

    def __init__(
        self,
        scheduler: FlowMatchEulerDiscreteScheduler,
        vae: AutoencoderKLQwenImage,
        text_encoder: Qwen2_5_VLForConditionalGeneration,
        tokenizer: Qwen2Tokenizer,
        transformer: QwenImageTransformer2DModel,
    ):
        super().__init__()

        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            tokenizer=tokenizer,
            transformer=transformer,
            scheduler=scheduler,
        )
        self.vae_scale_factor = 2 ** len(self.vae.temperal_downsample) if getattr(self, "vae", None) else 8
        # QwenImage latents are turned into 2x2 patches and packed. This means the latent width and height has to be divisible
        # by the patch size. So the vae scale factor is multiplied by the patch size to account for this
        self.latent_channels = self.vae.config.z_dim if getattr(self, "vae", None) else 16
        self.image_processor = VaeImageProcessor(
            vae_scale_factor=self.vae_scale_factor * 2, vae_latent_channels=self.latent_channels
        )
        self.mask_processor = VaeImageProcessor(
            vae_scale_factor=self.vae_scale_factor * 2,
            vae_latent_channels=self.latent_channels,
            do_normalize=False,
            do_binarize=True,
            do_convert_grayscale=True,
        )
        self.tokenizer_max_length = 1024
        self.prompt_template_encode = (
            "<|im_start|>system\nDescribe the image by detailing the color, shape, size, "
            "texture, quantity, text, spatial relationships of the objects and background:"
            "<|im_end|>\n<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
        )
        self.prompt_template_encode_start_idx = 34
        self.default_sample_size = 128

    # Copied from diffusers.pipelines.qwenimage.pipeline_qwenimage.QwenImagePipeline._extract_masked_hidden
    def _extract_masked_hidden(self, hidden_states: ms.Tensor, mask: ms.Tensor):
        bool_mask = mask.bool()
        valid_lengths = bool_mask.sum(dim=1)
        selected = hidden_states[bool_mask]
        split_result = mint.split(selected, valid_lengths.tolist(), dim=0)

        return split_result

    # Copied from diffusers.pipelines.qwenimage.pipeline_qwenimage.QwenImagePipeline._get_qwen_prompt_embeds
    def _get_qwen_prompt_embeds(
        self,
        prompt: Union[str, List[str]] = None,
        dtype: Optional[ms.Type] = None,
    ):
        dtype = dtype or self.text_encoder.dtype

        prompt = [prompt] if isinstance(prompt, str) else prompt

        template = self.prompt_template_encode
        drop_idx = self.prompt_template_encode_start_idx
        txt = [template.format(e) for e in prompt]
        txt_tokens = self.tokenizer(
            txt, max_length=self.tokenizer_max_length + drop_idx, padding=True, truncation=True, return_tensors="np"
        )
        encoder_hidden_states = self.text_encoder(
            input_ids=ms.tensor(txt_tokens.input_ids),
            attention_mask=ms.tensor(txt_tokens.attention_mask),
            output_hidden_states=True,
        )
        hidden_states = encoder_hidden_states.hidden_states[-1]
        split_hidden_states = self._extract_masked_hidden(hidden_states, ms.tensor(txt_tokens.attention_mask))
        split_hidden_states = [e[drop_idx:] for e in split_hidden_states]
        attn_mask_list = [mint.ones(e.shape[0], dtype=ms.int64) for e in split_hidden_states]
        max_seq_len = max([e.shape[0] for e in split_hidden_states])
        prompt_embeds = mint.stack(
            [mint.cat([u, u.new_zeros((max_seq_len - u.shape[0], u.shape[1]))]) for u in split_hidden_states]
        )
        encoder_attention_mask = mint.stack(
            [mint.cat([u, u.new_zeros((max_seq_len - u.shape[0]))]) for u in attn_mask_list]
        )

        prompt_embeds = prompt_embeds.to(dtype=dtype)

        return prompt_embeds, encoder_attention_mask

    # Copied from diffusers.pipelines.qwenimage.pipeline_qwenimage_img2img.QwenImageImg2ImgPipeline._encode_vae_image
    def _encode_vae_image(self, image: ms.Tensor, generator: np.random.Generator):
        if isinstance(generator, list):
            image_latents = [
                retrieve_latents(self.vae, self.vae.encode(image[i : i + 1])[0]) for i in range(image.shape[0])
            ]
            image_latents = mint.cat(image_latents, dim=0)
        else:
            image_latents = retrieve_latents(self.vae, self.vae.encode(image)[0])

        latents_mean = (
            ms.tensor(self.vae.config.latents_mean).view(1, self.vae.config.z_dim, 1, 1, 1).to(image_latents.dtype)
        )
        latents_std = 1.0 / ms.tensor(self.vae.config.latents_std).view(1, self.vae.config.z_dim, 1, 1, 1).to(
            image_latents.dtype
        )

        image_latents = (image_latents - latents_mean) * latents_std

        return image_latents

    # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img.StableDiffusion3Img2ImgPipeline.get_timesteps
    def get_timesteps(self, num_inference_steps, strength):
        # get the original timestep using init_timestep
        init_timestep = min(num_inference_steps * strength, num_inference_steps)

        t_start = int(max(num_inference_steps - init_timestep, 0))
        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
        if hasattr(self.scheduler, "set_begin_index"):
            self.scheduler.set_begin_index(t_start * self.scheduler.order)

        return timesteps, num_inference_steps - t_start

    # Copied fromCopied from diffusers.pipelines.qwenimage.pipeline_qwenimage.QwenImagePipeline.encode_prompt
    def encode_prompt(
        self,
        prompt: Union[str, List[str]],
        num_images_per_prompt: int = 1,
        prompt_embeds: Optional[ms.Tensor] = None,
        prompt_embeds_mask: Optional[ms.Tensor] = None,
        max_sequence_length: int = 1024,
    ):
        r"""

        Args:
            prompt (`str` or `List[str]`, *optional*):
                prompt to be encoded
            num_images_per_prompt (`int`):
                number of images that should be generated per prompt
            prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
        """
        prompt = [prompt] if isinstance(prompt, str) else prompt
        batch_size = len(prompt) if prompt_embeds is None else prompt_embeds.shape[0]

        if prompt_embeds is None:
            prompt_embeds, prompt_embeds_mask = self._get_qwen_prompt_embeds(prompt)

        prompt_embeds = prompt_embeds[:, :max_sequence_length]
        prompt_embeds_mask = prompt_embeds_mask[:, :max_sequence_length]

        _, seq_len, _ = prompt_embeds.shape
        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
        prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
        prompt_embeds_mask = prompt_embeds_mask.repeat(1, num_images_per_prompt, 1)
        prompt_embeds_mask = prompt_embeds_mask.view(batch_size * num_images_per_prompt, seq_len)

        return prompt_embeds, prompt_embeds_mask

    def check_inputs(
        self,
        prompt,
        image,
        mask_image,
        strength,
        height,
        width,
        output_type,
        negative_prompt=None,
        prompt_embeds=None,
        negative_prompt_embeds=None,
        prompt_embeds_mask=None,
        negative_prompt_embeds_mask=None,
        callback_on_step_end_tensor_inputs=None,
        padding_mask_crop=None,
        max_sequence_length=None,
    ):
        if strength < 0 or strength > 1:
            raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")

        if height % (self.vae_scale_factor * 2) != 0 or width % (self.vae_scale_factor * 2) != 0:
            logger.warning(
                f"`height` and `width` have to be divisible by {self.vae_scale_factor * 2} but are {height} and {width}."
                " Dimensions will be resized accordingly"
            )

        if callback_on_step_end_tensor_inputs is not None and not all(
            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
        ):
            raise ValueError(
                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found"
                f" {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
            )

        if prompt is not None and prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt is None and prompt_embeds is None:
            raise ValueError(
                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
            )
        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")

        if negative_prompt is not None and negative_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
            )

        if prompt_embeds is not None and prompt_embeds_mask is None:
            raise ValueError(
                "If `prompt_embeds` are provided, `prompt_embeds_mask` also have to be passed. Make sure to generate `prompt_embeds_mask`"
                " from the same text encoder that was used to generate `prompt_embeds`."
            )

        if negative_prompt_embeds is not None and negative_prompt_embeds_mask is None:
            raise ValueError(
                "If `negative_prompt_embeds` are provided, `negative_prompt_embeds_mask` also have to be passed. Make sure to generate"
                " `negative_prompt_embeds_mask` from the same text encoder that was used to generate `negative_prompt_embeds`."
            )

        if padding_mask_crop is not None:
            if not isinstance(image, PIL.Image.Image):
                raise ValueError(
                    f"The image should be a PIL image when inpainting mask crop, but is of type {type(image)}."
                )
            if not isinstance(mask_image, PIL.Image.Image):
                raise ValueError(
                    f"The mask image should be a PIL image when inpainting mask crop, but is of type"
                    f" {type(mask_image)}."
                )
            if output_type != "pil":
                raise ValueError(f"The output type should be PIL when inpainting mask crop, but is {output_type}.")

        if max_sequence_length is not None and max_sequence_length > 1024:
            raise ValueError(f"`max_sequence_length` cannot be greater than 1024 but is {max_sequence_length}")

    @staticmethod
    # Copied from diffusers.pipelines.qwenimage.pipeline_qwenimage.QwenImagePipeline._pack_latents
    def _pack_latents(latents, batch_size, num_channels_latents, height, width):
        latents = latents.view(batch_size, num_channels_latents, height // 2, 2, width // 2, 2)
        latents = latents.permute(0, 2, 4, 1, 3, 5)
        latents = latents.reshape(batch_size, (height // 2) * (width // 2), num_channels_latents * 4)

        return latents

    @staticmethod
    # Copied from diffusers.pipelines.qwenimage.pipeline_qwenimage.QwenImagePipeline._unpack_latents
    def _unpack_latents(latents, height, width, vae_scale_factor):
        batch_size, num_patches, channels = latents.shape

        # VAE applies 8x compression on images but we must also account for packing which requires
        # latent height and width to be divisible by 2.
        height = 2 * (int(height) // (vae_scale_factor * 2))
        width = 2 * (int(width) // (vae_scale_factor * 2))

        latents = latents.view(batch_size, height // 2, width // 2, channels // 4, 2, 2)
        latents = latents.permute(0, 3, 1, 4, 2, 5)

        latents = latents.reshape(batch_size, channels // (2 * 2), 1, height, width)

        return latents

    def enable_vae_slicing(self):
        r"""
        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
        """
        self.vae.enable_slicing()

    def disable_vae_slicing(self):
        r"""
        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
        computing decoding in one step.
        """
        self.vae.disable_slicing()

    def enable_vae_tiling(self):
        r"""
        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
        processing larger images.
        """
        self.vae.enable_tiling()

    def disable_vae_tiling(self):
        r"""
        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
        computing decoding in one step.
        """
        self.vae.disable_tiling()

    def prepare_latents(
        self,
        image,
        timestep,
        batch_size,
        num_channels_latents,
        height,
        width,
        dtype,
        generator,
        latents=None,
    ):
        if isinstance(generator, list) and len(generator) != batch_size:
            raise ValueError(
                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
            )
        # VAE applies 8x compression on images but we must also account for packing which requires
        # latent height and width to be divisible by 2.
        height = 2 * (int(height) // (self.vae_scale_factor * 2))
        width = 2 * (int(width) // (self.vae_scale_factor * 2))

        shape = (batch_size, 1, num_channels_latents, height, width)

        # If image is [B,C,H,W] -> add T=1. If it's already [B,C,T,H,W], leave it.
        if image.dim() == 4:
            image = image.unsqueeze(2)
        elif image.dim() != 5:
            raise ValueError(f"Expected image dims 4 or 5, got {image.dim()}.")

        if latents is not None:
            return latents.to(dtype=dtype)

        image = image.to(dtype=dtype)
        if image.shape[1] != self.latent_channels:
            image_latents = self._encode_vae_image(image=image, generator=generator)  # [B,z,1,H',W']
        else:
            image_latents = image
        if batch_size > image_latents.shape[0] and batch_size % image_latents.shape[0] == 0:
            # expand init_latents for batch_size
            additional_image_per_prompt = batch_size // image_latents.shape[0]
            image_latents = mint.cat([image_latents] * additional_image_per_prompt, dim=0)
        elif batch_size > image_latents.shape[0] and batch_size % image_latents.shape[0] != 0:
            raise ValueError(
                f"Cannot duplicate `image` of batch size {image_latents.shape[0]} to {batch_size} text prompts."
            )
        else:
            image_latents = mint.cat([image_latents], dim=0)

        image_latents = image_latents.transpose(1, 2)  # [B,1,z,H',W']

        if latents is None:
            noise = randn_tensor(shape, generator=generator, dtype=dtype)
            latents = self.scheduler.scale_noise(image_latents, timestep, noise)
        else:
            noise = latents
            latents = noise

        noise = self._pack_latents(noise, batch_size, num_channels_latents, height, width)
        image_latents = self._pack_latents(image_latents, batch_size, num_channels_latents, height, width)
        latents = self._pack_latents(latents, batch_size, num_channels_latents, height, width)

        return latents, noise, image_latents

    def prepare_mask_latents(
        self,
        mask,
        masked_image,
        batch_size,
        num_channels_latents,
        num_images_per_prompt,
        height,
        width,
        dtype,
        generator,
    ):
        # VAE applies 8x compression on images but we must also account for packing which requires
        # latent height and width to be divisible by 2.
        height = 2 * (int(height) // (self.vae_scale_factor * 2))
        width = 2 * (int(width) // (self.vae_scale_factor * 2))
        # resize the mask to latents shape as we concatenate the mask to the latents
        # we do that before converting to dtype to avoid breaking in case we're using cpu_offload
        # and half precision
        mask = mint.nn.functional.interpolate(mask, size=(height, width))
        mask = mask.to(dtype=dtype)

        batch_size = batch_size * num_images_per_prompt

        if masked_image.dim() == 4:
            masked_image = masked_image.unsqueeze(2)
        elif masked_image.dim() != 5:
            raise ValueError(f"Expected image dims 4 or 5, got {masked_image.dim()}.")

        masked_image = masked_image.to(dtype=dtype)

        if masked_image.shape[1] == self.latent_channels:
            masked_image_latents = masked_image
        else:
            masked_image_latents = self._encode_vae_image(image=masked_image, generator=generator)

            # duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method
        if mask.shape[0] < batch_size:
            if not batch_size % mask.shape[0] == 0:
                raise ValueError(
                    "The passed mask and the required batch size don't match. Masks are supposed to be duplicated to"
                    f" a total batch size of {batch_size}, but {mask.shape[0]} masks were passed. Make sure the number"
                    " of masks that you pass is divisible by the total requested batch size."
                )
            mask = mask.repeat(batch_size // mask.shape[0], 1, 1, 1)
        if masked_image_latents.shape[0] < batch_size:
            if not batch_size % masked_image_latents.shape[0] == 0:
                raise ValueError(
                    "The passed images and the required batch size don't match. Images are supposed to be duplicated"
                    f" to a total batch size of {batch_size}, but {masked_image_latents.shape[0]} images were passed."
                    " Make sure the number of images that you pass is divisible by the total requested batch size."
                )
            masked_image_latents = masked_image_latents.repeat(batch_size // masked_image_latents.shape[0], 1, 1, 1, 1)

        masked_image_latents = masked_image_latents.to(dtype=dtype)

        masked_image_latents = self._pack_latents(
            masked_image_latents,
            batch_size,
            num_channels_latents,
            height,
            width,
        )
        mask = self._pack_latents(
            mask.repeat(1, num_channels_latents, 1, 1),
            batch_size,
            num_channels_latents,
            height,
            width,
        )

        return mask, masked_image_latents

    @property
    def guidance_scale(self):
        return self._guidance_scale

    @property
    def attention_kwargs(self):
        return self._attention_kwargs

    @property
    def num_timesteps(self):
        return self._num_timesteps

    @property
    def current_timestep(self):
        return self._current_timestep

    @property
    def interrupt(self):
        return self._interrupt

    def __call__(
        self,
        prompt: Union[str, List[str]] = None,
        negative_prompt: Union[str, List[str]] = None,
        true_cfg_scale: float = 4.0,
        image: PipelineImageInput = None,
        mask_image: PipelineImageInput = None,
        masked_image_latents: PipelineImageInput = None,
        height: Optional[int] = None,
        width: Optional[int] = None,
        padding_mask_crop: Optional[int] = None,
        strength: float = 0.6,
        num_inference_steps: int = 50,
        sigmas: Optional[List[float]] = None,
        guidance_scale: Optional[float] = None,
        num_images_per_prompt: int = 1,
        generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
        latents: Optional[ms.Tensor] = None,
        prompt_embeds: Optional[ms.Tensor] = None,
        prompt_embeds_mask: Optional[ms.Tensor] = None,
        negative_prompt_embeds: Optional[ms.Tensor] = None,
        negative_prompt_embeds_mask: Optional[ms.Tensor] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = True,
        attention_kwargs: Optional[Dict[str, Any]] = None,
        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
        max_sequence_length: int = 512,
    ):
        r"""
        Function invoked when calling the pipeline for generation.

        Args:
            prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
                instead.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation. If not defined, one has to pass
                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
                not greater than `1`).
            image (`ms.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[ms.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`):
                `Image`, numpy array or tensor representing an image batch to be used as the starting point. For both
                numpy array and mindspore tensor, the expected value range is between `[0, 1]` If it's a tensor or a list
                or tensors, the expected shape should be `(B, C, H, W)` or `(C, H, W)`. If it is a numpy array or a
                list of arrays, the expected shape should be `(B, H, W, C)` or `(H, W, C)` It can also accept image
                latents as `image`, but if passing latents directly it is not encoded again.
            true_cfg_scale (`float`, *optional*, defaults to 1.0):
                Guidance scale as defined in [Classifier-Free Diffusion
                Guidance](https://huggingface.co/papers/2207.12598). `true_cfg_scale` is defined as `w` of equation 2.
                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Classifier-free guidance is enabled by
                setting `true_cfg_scale > 1` and a provided `negative_prompt`. Higher guidance scale encourages to
                generate images that are closely linked to the text `prompt`, usually at the expense of lower image
                quality.
            mask_image (`ms.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[ms.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`):
                `Image`, numpy array or tensor representing an image batch to mask `image`. White pixels in the mask
                are repainted while black pixels are preserved. If `mask_image` is a PIL image, it is converted to a
                single channel (luminance) before use. If it's a numpy array or mindspore tensor, it should contain one
                color channel (L) instead of 3, so the expected shape for mindspore tensor would be `(B, 1, H, W)`, `(B,
                H, W)`, `(1, H, W)`, `(H, W)`. And for numpy array would be for `(B, H, W, 1)`, `(B, H, W)`, `(H, W,
                1)`, or `(H, W)`.
            mask_image_latent (`ms.Tensor`, `List[ms.Tensor]`):
                `Tensor` representing an image batch to mask `image` generated by VAE. If not provided, the mask
                latents tensor will ge generated by `mask_image`.
            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                The height in pixels of the generated image. This is set to 1024 by default for the best results.
            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                The width in pixels of the generated image. This is set to 1024 by default for the best results.
            padding_mask_crop (`int`, *optional*, defaults to `None`):
                The size of margin in the crop to be applied to the image and masking. If `None`, no crop is applied to
                image and mask_image. If `padding_mask_crop` is not `None`, it will first find a rectangular region
                with the same aspect ration of the image and contains all masked area, and then expand that area based
                on `padding_mask_crop`. The image and mask_image will then be cropped based on the expanded area before
                resizing to the original image size for inpainting. This is useful when the masked area is small while
                the image is large and contain information irrelevant for inpainting, such as background.
            strength (`float`, *optional*, defaults to 1.0):
                Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a
                starting point and more noise is added the higher the `strength`. The number of denoising steps depends
                on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising
                process runs for the full number of iterations specified in `num_inference_steps`. A value of 1
                essentially ignores `image`.
            num_inference_steps (`int`, *optional*, defaults to 50):
                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                expense of slower inference.
            sigmas (`List[float]`, *optional*):
                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                will be used.
            guidance_scale (`float`, *optional*, defaults to 3.5):
                A guidance scale value for guidance distilled models. Unlike the traditional classifier-free guidance
                where the guidance scale is applied during inference through noise prediction rescaling, guidance
                distilled models take the guidance scale directly as an input parameter during forward pass. Guidance
                scale is enabled by setting `guidance_scale > 1`. Higher guidance scale encourages to generate images
                that are closely linked to the text `prompt`, usually at the expense of lower image quality. This
                parameter in the pipeline is there to support future guidance-distilled models when they come up. It is
                ignored when not using guidance distilled models. To enable traditional classifier-free guidance,
                please pass `true_cfg_scale > 1.0` and `negative_prompt` (even an empty negative prompt like " " should
                enable classifier-free guidance computations).
            num_images_per_prompt (`int`, *optional*, defaults to 1):
                The number of images to generate per prompt.
            generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
                One or a list of [numpy generator(s)](https://numpy.org/doc/stable/reference/random/generator.html)
                to make generation deterministic.
            latents (`ms.Tensor`, *optional*):
                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor will be generated by sampling using the supplied random `generator`.
            prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
            negative_prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                argument.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generate image. Choose between
                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.qwenimage.QwenImagePipelineOutput`] instead of a plain tuple.
            attention_kwargs (`dict`, *optional*):
                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
                `self.processor` in
                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
            callback_on_step_end (`Callable`, *optional*):
                A function that calls at the end of each denoising steps during the inference. The function is called
                with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
                callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
                `callback_on_step_end_tensor_inputs`.
            callback_on_step_end_tensor_inputs (`List`, *optional*):
                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
                `._callback_tensor_inputs` attribute of your pipeline class.
            max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.

        Examples:

        Returns:
            [`~pipelines.qwenimage.QwenImagePipelineOutput`] or `tuple`:
            [`~pipelines.qwenimage.QwenImagePipelineOutput`] if `return_dict` is True, otherwise a `tuple`. When
            returning a tuple, the first element is a list with the generated images.
        """

        height = height or self.default_sample_size * self.vae_scale_factor
        width = width or self.default_sample_size * self.vae_scale_factor

        # 1. Check inputs. Raise error if not correct
        self.check_inputs(
            prompt,
            image,
            mask_image,
            strength,
            height,
            width,
            output_type=output_type,
            negative_prompt=negative_prompt,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            prompt_embeds_mask=prompt_embeds_mask,
            negative_prompt_embeds_mask=negative_prompt_embeds_mask,
            callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
            padding_mask_crop=padding_mask_crop,
            max_sequence_length=max_sequence_length,
        )

        self._guidance_scale = guidance_scale
        self._attention_kwargs = attention_kwargs
        self._current_timestep = None
        self._interrupt = False

        # 2. Preprocess image
        if padding_mask_crop is not None:
            crops_coords = self.mask_processor.get_crop_region(mask_image, width, height, pad=padding_mask_crop)
            resize_mode = "fill"
        else:
            crops_coords = None
            resize_mode = "default"

        original_image = image
        init_image = self.image_processor.preprocess(
            image, height=height, width=width, crops_coords=crops_coords, resize_mode=resize_mode
        )
        init_image = init_image.to(dtype=ms.float32)

        # 3. Define call parameters
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        has_neg_prompt = negative_prompt is not None or (
            negative_prompt_embeds is not None and negative_prompt_embeds_mask is not None
        )

        if true_cfg_scale > 1 and not has_neg_prompt:
            logger.warning(
                f"true_cfg_scale is passed as {true_cfg_scale}, but classifier-free guidance is not enabled since no negative_prompt is provided."
            )
        elif true_cfg_scale <= 1 and has_neg_prompt:
            logger.warning(
                " negative_prompt is passed but classifier-free guidance is not enabled since true_cfg_scale <= 1"
            )

        do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
        prompt_embeds, prompt_embeds_mask = self.encode_prompt(
            prompt=prompt,
            prompt_embeds=prompt_embeds,
            prompt_embeds_mask=prompt_embeds_mask,
            num_images_per_prompt=num_images_per_prompt,
            max_sequence_length=max_sequence_length,
        )
        if do_true_cfg:
            negative_prompt_embeds, negative_prompt_embeds_mask = self.encode_prompt(
                prompt=negative_prompt,
                prompt_embeds=negative_prompt_embeds,
                prompt_embeds_mask=negative_prompt_embeds_mask,
                num_images_per_prompt=num_images_per_prompt,
                max_sequence_length=max_sequence_length,
            )

        # 4. Prepare timesteps
        sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else sigmas
        image_seq_len = (int(height) // self.vae_scale_factor // 2) * (int(width) // self.vae_scale_factor // 2)
        mu = calculate_shift(
            image_seq_len,
            self.scheduler.config.get("base_image_seq_len", 256),
            self.scheduler.config.get("max_image_seq_len", 4096),
            self.scheduler.config.get("base_shift", 0.5),
            self.scheduler.config.get("max_shift", 1.15),
        )
        timesteps, num_inference_steps = retrieve_timesteps(
            self.scheduler,
            num_inference_steps,
            sigmas=sigmas,
            mu=mu,
        )
        timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength)

        if num_inference_steps < 1:
            raise ValueError(
                f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline"
                f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline."
            )
        latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)

        # 5. Prepare latent variables
        num_channels_latents = self.transformer.config.in_channels // 4

        latents, noise, image_latents = self.prepare_latents(
            init_image,
            latent_timestep,
            batch_size * num_images_per_prompt,
            num_channels_latents,
            height,
            width,
            prompt_embeds.dtype,
            generator,
            latents,
        )

        mask_condition = self.mask_processor.preprocess(
            mask_image, height=height, width=width, resize_mode=resize_mode, crops_coords=crops_coords
        )

        if masked_image_latents is None:
            masked_image = init_image * (mask_condition < 0.5)
        else:
            masked_image = masked_image_latents

        mask, masked_image_latents = self.prepare_mask_latents(
            mask_condition,
            masked_image,
            batch_size,
            num_channels_latents,
            num_images_per_prompt,
            height,
            width,
            prompt_embeds.dtype,
            generator,
        )

        img_shapes = [[(1, height // self.vae_scale_factor // 2, width // self.vae_scale_factor // 2)]] * batch_size

        num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
        self._num_timesteps = len(timesteps)

        # handle guidance
        if self.transformer.config.guidance_embeds and guidance_scale is None:
            raise ValueError("guidance_scale is required for guidance-distilled model.")
        elif self.transformer.config.guidance_embeds:
            guidance = mint.full([1], guidance_scale, dtype=ms.float32)
            guidance = guidance.expand((latents.shape[0],))
        elif not self.transformer.config.guidance_embeds and guidance_scale is not None:
            logger.warning(
                f"guidance_scale is passed as {guidance_scale}, but ignored since the model is not guidance-distilled."
            )
            guidance = None
        elif not self.transformer.config.guidance_embeds and guidance_scale is None:
            guidance = None

        if self.attention_kwargs is None:
            self._attention_kwargs = {}

        txt_seq_lens = prompt_embeds_mask.sum(dim=1).tolist() if prompt_embeds_mask is not None else None
        negative_txt_seq_lens = (
            negative_prompt_embeds_mask.sum(dim=1).tolist() if negative_prompt_embeds_mask is not None else None
        )

        # 6. Denoising loop
        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(timesteps):
                if self.interrupt:
                    continue

                self._current_timestep = t
                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
                timestep = t.expand((latents.shape[0],)).to(latents.dtype)
                noise_pred = self.transformer(
                    hidden_states=latents,
                    timestep=timestep / 1000,
                    guidance=guidance,
                    encoder_hidden_states_mask=prompt_embeds_mask,
                    encoder_hidden_states=prompt_embeds,
                    img_shapes=img_shapes,
                    txt_seq_lens=txt_seq_lens,
                    attention_kwargs=self.attention_kwargs,
                    return_dict=False,
                )[0]

                if do_true_cfg:
                    neg_noise_pred = self.transformer(
                        hidden_states=latents,
                        timestep=timestep / 1000,
                        guidance=guidance,
                        encoder_hidden_states_mask=negative_prompt_embeds_mask,
                        encoder_hidden_states=negative_prompt_embeds,
                        img_shapes=img_shapes,
                        txt_seq_lens=negative_txt_seq_lens,
                        attention_kwargs=self.attention_kwargs,
                        return_dict=False,
                    )[0]
                    comb_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)

                    cond_norm = mint.norm(noise_pred, dim=-1, keepdim=True)
                    noise_norm = mint.norm(comb_pred, dim=-1, keepdim=True)
                    noise_pred = comb_pred * (cond_norm / noise_norm)

                # compute the previous noisy sample x_t -> x_t-1
                latents_dtype = latents.dtype
                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]

                # for 64 channel transformer only.
                init_latents_proper = image_latents
                init_mask = mask

                if i < len(timesteps) - 1:
                    noise_timestep = timesteps[i + 1]
                    init_latents_proper = self.scheduler.scale_noise(
                        init_latents_proper, ms.tensor([noise_timestep]), noise
                    )

                latents = (1 - init_mask) * init_latents_proper + init_mask * latents

                if latents.dtype != latents_dtype:
                    latents = latents.to(latents_dtype)

                if callback_on_step_end is not None:
                    callback_kwargs = {}
                    for k in callback_on_step_end_tensor_inputs:
                        callback_kwargs[k] = locals()[k]
                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)

                    latents = callback_outputs.pop("latents", latents)
                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)

                # call the callback, if provided
                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                    progress_bar.update()

        self._current_timestep = None
        if output_type == "latent":
            image = latents
        else:
            latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
            latents = latents.to(self.vae.dtype)
            latents_mean = (
                ms.tensor(self.vae.config.latents_mean).view(1, self.vae.config.z_dim, 1, 1, 1).to(latents.dtype)
            )
            latents_std = 1.0 / ms.tensor(self.vae.config.latents_std).view(1, self.vae.config.z_dim, 1, 1, 1).to(
                latents.dtype
            )

            latents = latents / latents_std + latents_mean
            image = self.vae.decode(latents, return_dict=False)[0][:, :, 0]
            image = self.image_processor.postprocess(image, output_type=output_type)

            if padding_mask_crop is not None:
                image = [self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image]

        if not return_dict:
            return (image,)

        return QwenImagePipelineOutput(images=image)

mindone.diffusers.QwenImageInpaintPipeline.__call__(prompt=None, negative_prompt=None, true_cfg_scale=4.0, image=None, mask_image=None, masked_image_latents=None, height=None, width=None, padding_mask_crop=None, strength=0.6, num_inference_steps=50, sigmas=None, guidance_scale=None, num_images_per_prompt=1, generator=None, latents=None, prompt_embeds=None, prompt_embeds_mask=None, negative_prompt_embeds=None, negative_prompt_embeds_mask=None, output_type='pil', return_dict=True, attention_kwargs=None, callback_on_step_end=None, callback_on_step_end_tensor_inputs=['latents'], max_sequence_length=512)

Function invoked when calling the pipeline for generation.

PARAMETER	DESCRIPTION
prompt	The prompt or prompts to guide the image generation. If not defined, one has to pass prompt_embeds. instead. TYPE: `str` or `List[str]`, *optional* DEFAULT: None
negative_prompt	The prompt or prompts not to guide the image generation. If not defined, one has to pass negative_prompt_embeds instead. Ignored when not using guidance (i.e., ignored if true_cfg_scale is not greater than 1). TYPE: `str` or `List[str]`, *optional* DEFAULT: None
image	Image, numpy array or tensor representing an image batch to be used as the starting point. For both numpy array and mindspore tensor, the expected value range is between [0, 1] If it's a tensor or a list or tensors, the expected shape should be (B, C, H, W) or (C, H, W). If it is a numpy array or a list of arrays, the expected shape should be (B, H, W, C) or (H, W, C) It can also accept image latents as image, but if passing latents directly it is not encoded again. TYPE: `ms.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[ms.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]` DEFAULT: None
true_cfg_scale	Guidance scale as defined in Classifier-Free Diffusion Guidance. true_cfg_scale is defined as w of equation 2. of Imagen Paper. Classifier-free guidance is enabled by setting true_cfg_scale > 1 and a provided negative_prompt. Higher guidance scale encourages to generate images that are closely linked to the text prompt, usually at the expense of lower image quality. TYPE: `float`, *optional*, defaults to 1.0 DEFAULT: 4.0
mask_image	Image, numpy array or tensor representing an image batch to mask image. White pixels in the mask are repainted while black pixels are preserved. If mask_image is a PIL image, it is converted to a single channel (luminance) before use. If it's a numpy array or mindspore tensor, it should contain one color channel (L) instead of 3, so the expected shape for mindspore tensor would be (B, 1, H, W), (B, H, W), (1, H, W), (H, W). And for numpy array would be for (B, H, W, 1), (B, H, W), (H, W, 1), or (H, W). TYPE: `ms.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[ms.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]` DEFAULT: None
mask_image_latent	Tensor representing an image batch to mask image generated by VAE. If not provided, the mask latents tensor will ge generated by mask_image. TYPE: `ms.Tensor`, `List[ms.Tensor]`
height	The height in pixels of the generated image. This is set to 1024 by default for the best results. TYPE: `int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor DEFAULT: None
width	The width in pixels of the generated image. This is set to 1024 by default for the best results. TYPE: `int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor DEFAULT: None
padding_mask_crop	The size of margin in the crop to be applied to the image and masking. If None, no crop is applied to image and mask_image. If padding_mask_crop is not None, it will first find a rectangular region with the same aspect ration of the image and contains all masked area, and then expand that area based on padding_mask_crop. The image and mask_image will then be cropped based on the expanded area before resizing to the original image size for inpainting. This is useful when the masked area is small while the image is large and contain information irrelevant for inpainting, such as background. TYPE: `int`, *optional*, defaults to `None` DEFAULT: None
strength	Indicates extent to transform the reference image. Must be between 0 and 1. image is used as a starting point and more noise is added the higher the strength. The number of denoising steps depends on the amount of noise initially added. When strength is 1, added noise is maximum and the denoising process runs for the full number of iterations specified in num_inference_steps. A value of 1 essentially ignores image. TYPE: `float`, *optional*, defaults to 1.0 DEFAULT: 0.6
num_inference_steps	The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. TYPE: `int`, *optional*, defaults to 50 DEFAULT: 50
sigmas	Custom sigmas to use for the denoising process with schedulers which support a sigmas argument in their set_timesteps method. If not defined, the default behavior when num_inference_steps is passed will be used. TYPE: `List[float]`, *optional* DEFAULT: None
guidance_scale	A guidance scale value for guidance distilled models. Unlike the traditional classifier-free guidance where the guidance scale is applied during inference through noise prediction rescaling, guidance distilled models take the guidance scale directly as an input parameter during forward pass. Guidance scale is enabled by setting guidance_scale > 1. Higher guidance scale encourages to generate images that are closely linked to the text prompt, usually at the expense of lower image quality. This parameter in the pipeline is there to support future guidance-distilled models when they come up. It is ignored when not using guidance distilled models. To enable traditional classifier-free guidance, please pass true_cfg_scale > 1.0 and negative_prompt (even an empty negative prompt like " " should enable classifier-free guidance computations). TYPE: `float`, *optional*, defaults to 3.5 DEFAULT: None
num_images_per_prompt	The number of images to generate per prompt. TYPE: `int`, *optional*, defaults to 1 DEFAULT: 1
generator	One or a list of numpy generator(s) to make generation deterministic. TYPE: `np.random.Generator` or `List[np.random.Generator]`, *optional* DEFAULT: None
latents	Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor will be generated by sampling using the supplied random generator. TYPE: `ms.Tensor`, *optional* DEFAULT: None
prompt_embeds	Pre-generated text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, text embeddings will be generated from prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
negative_prompt_embeds	Pre-generated negative text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, negative_prompt_embeds will be generated from negative_prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
output_type	The output format of the generate image. Choose between PIL: PIL.Image.Image or np.array. TYPE: `str`, *optional*, defaults to `"pil"` DEFAULT: 'pil'
return_dict	Whether or not to return a [~pipelines.qwenimage.QwenImagePipelineOutput] instead of a plain tuple. TYPE: `bool`, *optional*, defaults to `True` DEFAULT: True
attention_kwargs	A kwargs dictionary that if specified is passed along to the AttentionProcessor as defined under self.processor in diffusers.models.attention_processor. TYPE: `dict`, *optional* DEFAULT: None
callback_on_step_end	A function that calls at the end of each denoising steps during the inference. The function is called with the following arguments: callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict). callback_kwargs will include a list of all tensors as specified by callback_on_step_end_tensor_inputs. TYPE: `Callable`, *optional* DEFAULT: None
callback_on_step_end_tensor_inputs	The list of tensor inputs for the callback_on_step_end function. The tensors specified in the list will be passed as callback_kwargs argument. You will only be able to include variables listed in the ._callback_tensor_inputs attribute of your pipeline class. TYPE: `List`, *optional* DEFAULT: ['latents']
max_sequence_length	Maximum sequence length to use with the prompt. TYPE: `int` defaults to 512 DEFAULT: 512

RETURNS	DESCRIPTION
	[~pipelines.qwenimage.QwenImagePipelineOutput] or tuple:
	[~pipelines.qwenimage.QwenImagePipelineOutput] if return_dict is True, otherwise a tuple. When
	returning a tuple, the first element is a list with the generated images.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py

def __call__(
    self,
    prompt: Union[str, List[str]] = None,
    negative_prompt: Union[str, List[str]] = None,
    true_cfg_scale: float = 4.0,
    image: PipelineImageInput = None,
    mask_image: PipelineImageInput = None,
    masked_image_latents: PipelineImageInput = None,
    height: Optional[int] = None,
    width: Optional[int] = None,
    padding_mask_crop: Optional[int] = None,
    strength: float = 0.6,
    num_inference_steps: int = 50,
    sigmas: Optional[List[float]] = None,
    guidance_scale: Optional[float] = None,
    num_images_per_prompt: int = 1,
    generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
    latents: Optional[ms.Tensor] = None,
    prompt_embeds: Optional[ms.Tensor] = None,
    prompt_embeds_mask: Optional[ms.Tensor] = None,
    negative_prompt_embeds: Optional[ms.Tensor] = None,
    negative_prompt_embeds_mask: Optional[ms.Tensor] = None,
    output_type: Optional[str] = "pil",
    return_dict: bool = True,
    attention_kwargs: Optional[Dict[str, Any]] = None,
    callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
    callback_on_step_end_tensor_inputs: List[str] = ["latents"],
    max_sequence_length: int = 512,
):
    r"""
    Function invoked when calling the pipeline for generation.

    Args:
        prompt (`str` or `List[str]`, *optional*):
            The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
            instead.
        negative_prompt (`str` or `List[str]`, *optional*):
            The prompt or prompts not to guide the image generation. If not defined, one has to pass
            `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
            not greater than `1`).
        image (`ms.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[ms.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`):
            `Image`, numpy array or tensor representing an image batch to be used as the starting point. For both
            numpy array and mindspore tensor, the expected value range is between `[0, 1]` If it's a tensor or a list
            or tensors, the expected shape should be `(B, C, H, W)` or `(C, H, W)`. If it is a numpy array or a
            list of arrays, the expected shape should be `(B, H, W, C)` or `(H, W, C)` It can also accept image
            latents as `image`, but if passing latents directly it is not encoded again.
        true_cfg_scale (`float`, *optional*, defaults to 1.0):
            Guidance scale as defined in [Classifier-Free Diffusion
            Guidance](https://huggingface.co/papers/2207.12598). `true_cfg_scale` is defined as `w` of equation 2.
            of [Imagen Paper](https://huggingface.co/papers/2205.11487). Classifier-free guidance is enabled by
            setting `true_cfg_scale > 1` and a provided `negative_prompt`. Higher guidance scale encourages to
            generate images that are closely linked to the text `prompt`, usually at the expense of lower image
            quality.
        mask_image (`ms.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[ms.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`):
            `Image`, numpy array or tensor representing an image batch to mask `image`. White pixels in the mask
            are repainted while black pixels are preserved. If `mask_image` is a PIL image, it is converted to a
            single channel (luminance) before use. If it's a numpy array or mindspore tensor, it should contain one
            color channel (L) instead of 3, so the expected shape for mindspore tensor would be `(B, 1, H, W)`, `(B,
            H, W)`, `(1, H, W)`, `(H, W)`. And for numpy array would be for `(B, H, W, 1)`, `(B, H, W)`, `(H, W,
            1)`, or `(H, W)`.
        mask_image_latent (`ms.Tensor`, `List[ms.Tensor]`):
            `Tensor` representing an image batch to mask `image` generated by VAE. If not provided, the mask
            latents tensor will ge generated by `mask_image`.
        height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
            The height in pixels of the generated image. This is set to 1024 by default for the best results.
        width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
            The width in pixels of the generated image. This is set to 1024 by default for the best results.
        padding_mask_crop (`int`, *optional*, defaults to `None`):
            The size of margin in the crop to be applied to the image and masking. If `None`, no crop is applied to
            image and mask_image. If `padding_mask_crop` is not `None`, it will first find a rectangular region
            with the same aspect ration of the image and contains all masked area, and then expand that area based
            on `padding_mask_crop`. The image and mask_image will then be cropped based on the expanded area before
            resizing to the original image size for inpainting. This is useful when the masked area is small while
            the image is large and contain information irrelevant for inpainting, such as background.
        strength (`float`, *optional*, defaults to 1.0):
            Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a
            starting point and more noise is added the higher the `strength`. The number of denoising steps depends
            on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising
            process runs for the full number of iterations specified in `num_inference_steps`. A value of 1
            essentially ignores `image`.
        num_inference_steps (`int`, *optional*, defaults to 50):
            The number of denoising steps. More denoising steps usually lead to a higher quality image at the
            expense of slower inference.
        sigmas (`List[float]`, *optional*):
            Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
            their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
            will be used.
        guidance_scale (`float`, *optional*, defaults to 3.5):
            A guidance scale value for guidance distilled models. Unlike the traditional classifier-free guidance
            where the guidance scale is applied during inference through noise prediction rescaling, guidance
            distilled models take the guidance scale directly as an input parameter during forward pass. Guidance
            scale is enabled by setting `guidance_scale > 1`. Higher guidance scale encourages to generate images
            that are closely linked to the text `prompt`, usually at the expense of lower image quality. This
            parameter in the pipeline is there to support future guidance-distilled models when they come up. It is
            ignored when not using guidance distilled models. To enable traditional classifier-free guidance,
            please pass `true_cfg_scale > 1.0` and `negative_prompt` (even an empty negative prompt like " " should
            enable classifier-free guidance computations).
        num_images_per_prompt (`int`, *optional*, defaults to 1):
            The number of images to generate per prompt.
        generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
            One or a list of [numpy generator(s)](https://numpy.org/doc/stable/reference/random/generator.html)
            to make generation deterministic.
        latents (`ms.Tensor`, *optional*):
            Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
            generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
            tensor will be generated by sampling using the supplied random `generator`.
        prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
            provided, text embeddings will be generated from `prompt` input argument.
        negative_prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
            weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
            argument.
        output_type (`str`, *optional*, defaults to `"pil"`):
            The output format of the generate image. Choose between
            [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
        return_dict (`bool`, *optional*, defaults to `True`):
            Whether or not to return a [`~pipelines.qwenimage.QwenImagePipelineOutput`] instead of a plain tuple.
        attention_kwargs (`dict`, *optional*):
            A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
            `self.processor` in
            [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
        callback_on_step_end (`Callable`, *optional*):
            A function that calls at the end of each denoising steps during the inference. The function is called
            with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
            callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
            `callback_on_step_end_tensor_inputs`.
        callback_on_step_end_tensor_inputs (`List`, *optional*):
            The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
            will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
            `._callback_tensor_inputs` attribute of your pipeline class.
        max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.

    Examples:

    Returns:
        [`~pipelines.qwenimage.QwenImagePipelineOutput`] or `tuple`:
        [`~pipelines.qwenimage.QwenImagePipelineOutput`] if `return_dict` is True, otherwise a `tuple`. When
        returning a tuple, the first element is a list with the generated images.
    """

    height = height or self.default_sample_size * self.vae_scale_factor
    width = width or self.default_sample_size * self.vae_scale_factor

    # 1. Check inputs. Raise error if not correct
    self.check_inputs(
        prompt,
        image,
        mask_image,
        strength,
        height,
        width,
        output_type=output_type,
        negative_prompt=negative_prompt,
        prompt_embeds=prompt_embeds,
        negative_prompt_embeds=negative_prompt_embeds,
        prompt_embeds_mask=prompt_embeds_mask,
        negative_prompt_embeds_mask=negative_prompt_embeds_mask,
        callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
        padding_mask_crop=padding_mask_crop,
        max_sequence_length=max_sequence_length,
    )

    self._guidance_scale = guidance_scale
    self._attention_kwargs = attention_kwargs
    self._current_timestep = None
    self._interrupt = False

    # 2. Preprocess image
    if padding_mask_crop is not None:
        crops_coords = self.mask_processor.get_crop_region(mask_image, width, height, pad=padding_mask_crop)
        resize_mode = "fill"
    else:
        crops_coords = None
        resize_mode = "default"

    original_image = image
    init_image = self.image_processor.preprocess(
        image, height=height, width=width, crops_coords=crops_coords, resize_mode=resize_mode
    )
    init_image = init_image.to(dtype=ms.float32)

    # 3. Define call parameters
    if prompt is not None and isinstance(prompt, str):
        batch_size = 1
    elif prompt is not None and isinstance(prompt, list):
        batch_size = len(prompt)
    else:
        batch_size = prompt_embeds.shape[0]

    has_neg_prompt = negative_prompt is not None or (
        negative_prompt_embeds is not None and negative_prompt_embeds_mask is not None
    )

    if true_cfg_scale > 1 and not has_neg_prompt:
        logger.warning(
            f"true_cfg_scale is passed as {true_cfg_scale}, but classifier-free guidance is not enabled since no negative_prompt is provided."
        )
    elif true_cfg_scale <= 1 and has_neg_prompt:
        logger.warning(
            " negative_prompt is passed but classifier-free guidance is not enabled since true_cfg_scale <= 1"
        )

    do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
    prompt_embeds, prompt_embeds_mask = self.encode_prompt(
        prompt=prompt,
        prompt_embeds=prompt_embeds,
        prompt_embeds_mask=prompt_embeds_mask,
        num_images_per_prompt=num_images_per_prompt,
        max_sequence_length=max_sequence_length,
    )
    if do_true_cfg:
        negative_prompt_embeds, negative_prompt_embeds_mask = self.encode_prompt(
            prompt=negative_prompt,
            prompt_embeds=negative_prompt_embeds,
            prompt_embeds_mask=negative_prompt_embeds_mask,
            num_images_per_prompt=num_images_per_prompt,
            max_sequence_length=max_sequence_length,
        )

    # 4. Prepare timesteps
    sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else sigmas
    image_seq_len = (int(height) // self.vae_scale_factor // 2) * (int(width) // self.vae_scale_factor // 2)
    mu = calculate_shift(
        image_seq_len,
        self.scheduler.config.get("base_image_seq_len", 256),
        self.scheduler.config.get("max_image_seq_len", 4096),
        self.scheduler.config.get("base_shift", 0.5),
        self.scheduler.config.get("max_shift", 1.15),
    )
    timesteps, num_inference_steps = retrieve_timesteps(
        self.scheduler,
        num_inference_steps,
        sigmas=sigmas,
        mu=mu,
    )
    timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength)

    if num_inference_steps < 1:
        raise ValueError(
            f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline"
            f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline."
        )
    latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)

    # 5. Prepare latent variables
    num_channels_latents = self.transformer.config.in_channels // 4

    latents, noise, image_latents = self.prepare_latents(
        init_image,
        latent_timestep,
        batch_size * num_images_per_prompt,
        num_channels_latents,
        height,
        width,
        prompt_embeds.dtype,
        generator,
        latents,
    )

    mask_condition = self.mask_processor.preprocess(
        mask_image, height=height, width=width, resize_mode=resize_mode, crops_coords=crops_coords
    )

    if masked_image_latents is None:
        masked_image = init_image * (mask_condition < 0.5)
    else:
        masked_image = masked_image_latents

    mask, masked_image_latents = self.prepare_mask_latents(
        mask_condition,
        masked_image,
        batch_size,
        num_channels_latents,
        num_images_per_prompt,
        height,
        width,
        prompt_embeds.dtype,
        generator,
    )

    img_shapes = [[(1, height // self.vae_scale_factor // 2, width // self.vae_scale_factor // 2)]] * batch_size

    num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
    self._num_timesteps = len(timesteps)

    # handle guidance
    if self.transformer.config.guidance_embeds and guidance_scale is None:
        raise ValueError("guidance_scale is required for guidance-distilled model.")
    elif self.transformer.config.guidance_embeds:
        guidance = mint.full([1], guidance_scale, dtype=ms.float32)
        guidance = guidance.expand((latents.shape[0],))
    elif not self.transformer.config.guidance_embeds and guidance_scale is not None:
        logger.warning(
            f"guidance_scale is passed as {guidance_scale}, but ignored since the model is not guidance-distilled."
        )
        guidance = None
    elif not self.transformer.config.guidance_embeds and guidance_scale is None:
        guidance = None

    if self.attention_kwargs is None:
        self._attention_kwargs = {}

    txt_seq_lens = prompt_embeds_mask.sum(dim=1).tolist() if prompt_embeds_mask is not None else None
    negative_txt_seq_lens = (
        negative_prompt_embeds_mask.sum(dim=1).tolist() if negative_prompt_embeds_mask is not None else None
    )

    # 6. Denoising loop
    with self.progress_bar(total=num_inference_steps) as progress_bar:
        for i, t in enumerate(timesteps):
            if self.interrupt:
                continue

            self._current_timestep = t
            # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
            timestep = t.expand((latents.shape[0],)).to(latents.dtype)
            noise_pred = self.transformer(
                hidden_states=latents,
                timestep=timestep / 1000,
                guidance=guidance,
                encoder_hidden_states_mask=prompt_embeds_mask,
                encoder_hidden_states=prompt_embeds,
                img_shapes=img_shapes,
                txt_seq_lens=txt_seq_lens,
                attention_kwargs=self.attention_kwargs,
                return_dict=False,
            )[0]

            if do_true_cfg:
                neg_noise_pred = self.transformer(
                    hidden_states=latents,
                    timestep=timestep / 1000,
                    guidance=guidance,
                    encoder_hidden_states_mask=negative_prompt_embeds_mask,
                    encoder_hidden_states=negative_prompt_embeds,
                    img_shapes=img_shapes,
                    txt_seq_lens=negative_txt_seq_lens,
                    attention_kwargs=self.attention_kwargs,
                    return_dict=False,
                )[0]
                comb_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)

                cond_norm = mint.norm(noise_pred, dim=-1, keepdim=True)
                noise_norm = mint.norm(comb_pred, dim=-1, keepdim=True)
                noise_pred = comb_pred * (cond_norm / noise_norm)

            # compute the previous noisy sample x_t -> x_t-1
            latents_dtype = latents.dtype
            latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]

            # for 64 channel transformer only.
            init_latents_proper = image_latents
            init_mask = mask

            if i < len(timesteps) - 1:
                noise_timestep = timesteps[i + 1]
                init_latents_proper = self.scheduler.scale_noise(
                    init_latents_proper, ms.tensor([noise_timestep]), noise
                )

            latents = (1 - init_mask) * init_latents_proper + init_mask * latents

            if latents.dtype != latents_dtype:
                latents = latents.to(latents_dtype)

            if callback_on_step_end is not None:
                callback_kwargs = {}
                for k in callback_on_step_end_tensor_inputs:
                    callback_kwargs[k] = locals()[k]
                callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)

                latents = callback_outputs.pop("latents", latents)
                prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)

            # call the callback, if provided
            if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                progress_bar.update()

    self._current_timestep = None
    if output_type == "latent":
        image = latents
    else:
        latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
        latents = latents.to(self.vae.dtype)
        latents_mean = (
            ms.tensor(self.vae.config.latents_mean).view(1, self.vae.config.z_dim, 1, 1, 1).to(latents.dtype)
        )
        latents_std = 1.0 / ms.tensor(self.vae.config.latents_std).view(1, self.vae.config.z_dim, 1, 1, 1).to(
            latents.dtype
        )

        latents = latents / latents_std + latents_mean
        image = self.vae.decode(latents, return_dict=False)[0][:, :, 0]
        image = self.image_processor.postprocess(image, output_type=output_type)

        if padding_mask_crop is not None:
            image = [self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image]

    if not return_dict:
        return (image,)

    return QwenImagePipelineOutput(images=image)

mindone.diffusers.QwenImageInpaintPipeline.disable_vae_slicing()

Disable sliced VAE decoding. If enable_vae_slicing was previously enabled, this method will go back to computing decoding in one step.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py

def disable_vae_slicing(self):
    r"""
    Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
    computing decoding in one step.
    """
    self.vae.disable_slicing()

mindone.diffusers.QwenImageInpaintPipeline.disable_vae_tiling()

Disable tiled VAE decoding. If enable_vae_tiling was previously enabled, this method will go back to computing decoding in one step.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py

def disable_vae_tiling(self):
    r"""
    Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
    computing decoding in one step.
    """
    self.vae.disable_tiling()

mindone.diffusers.QwenImageInpaintPipeline.enable_vae_slicing()

Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py

def enable_vae_slicing(self):
    r"""
    Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
    compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
    """
    self.vae.enable_slicing()

mindone.diffusers.QwenImageInpaintPipeline.enable_vae_tiling()

Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow processing larger images.

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py

def enable_vae_tiling(self):
    r"""
    Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
    compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
    processing larger images.
    """
    self.vae.enable_tiling()

mindone.diffusers.QwenImageInpaintPipeline.encode_prompt(prompt, num_images_per_prompt=1, prompt_embeds=None, prompt_embeds_mask=None, max_sequence_length=1024)

PARAMETER	DESCRIPTION
prompt	prompt to be encoded TYPE: `str` or `List[str]`, *optional*
num_images_per_prompt	number of images that should be generated per prompt TYPE: `int` DEFAULT: 1
prompt_embeds	Pre-generated text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, text embeddings will be generated from prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None

Source code in mindone/diffusers/pipelines/qwenimage/pipeline_qwenimage_inpaint.py

def encode_prompt(
    self,
    prompt: Union[str, List[str]],
    num_images_per_prompt: int = 1,
    prompt_embeds: Optional[ms.Tensor] = None,
    prompt_embeds_mask: Optional[ms.Tensor] = None,
    max_sequence_length: int = 1024,
):
    r"""

    Args:
        prompt (`str` or `List[str]`, *optional*):
            prompt to be encoded
        num_images_per_prompt (`int`):
            number of images that should be generated per prompt
        prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
            provided, text embeddings will be generated from `prompt` input argument.
    """
    prompt = [prompt] if isinstance(prompt, str) else prompt
    batch_size = len(prompt) if prompt_embeds is None else prompt_embeds.shape[0]

    if prompt_embeds is None:
        prompt_embeds, prompt_embeds_mask = self._get_qwen_prompt_embeds(prompt)

    prompt_embeds = prompt_embeds[:, :max_sequence_length]
    prompt_embeds_mask = prompt_embeds_mask[:, :max_sequence_length]

    _, seq_len, _ = prompt_embeds.shape
    prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
    prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
    prompt_embeds_mask = prompt_embeds_mask.repeat(1, num_images_per_prompt, 1)
    prompt_embeds_mask = prompt_embeds_mask.view(batch_size * num_images_per_prompt, seq_len)

    return prompt_embeds, prompt_embeds_mask