Stable Diffusion 3

Stable Diffusion 3 (SD3) was proposed in Scaling Rectified Flow Transformers for High-Resolution Image Synthesis by Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Muller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, Dustin Podell, Tim Dockhorn, Zion English, Kyle Lacey, Alex Goodwin, Yannik Marek, and Robin Rombach.

The abstract from the paper is:

Diffusion models create data from noise by inverting the forward paths of data towards noise and have emerged as a powerful generative modeling technique for high-dimensional, perceptual data such as images and videos. Rectified flow is a recent generative model formulation that connects data and noise in a straight line. Despite its better theoretical properties and conceptual simplicity, it is not yet decisively established as standard practice. In this work, we improve existing noise sampling techniques for training rectified flow models by biasing them towards perceptually relevant scales. Through a large-scale study, we demonstrate the superior performance of this approach compared to established diffusion formulations for high-resolution text-to-image synthesis. Additionally, we present a novel transformer-based architecture for text-to-image generation that uses separate weights for the two modalities and enables a bidirectional flow of information between image and text tokens, improving text comprehension typography, and human preference ratings. We demonstrate that this architecture follows predictable scaling trends and correlates lower validation loss to improved text-to-image synthesis as measured by various metrics and human evaluations.

Usage Example

As the model is gated, before using it with diffusers you first need to go to the Stable Diffusion 3 Medium Hugging Face page, fill in the form and accept the gate. Once you are in, you need to login so that your system knows you’ve accepted the gate.

Use the command below to log in:

huggingface-cli login

Tip

The SD3 pipeline uses three text encoders to generate an image. Model offloading is necessary in order for it to run on most commodity hardware. Please use the ms.float16 data type for additional memory savings.

import mindspore as ms
from mindone.diffusers import StableDiffusion3Pipeline

pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3-medium-diffusers", mindspore_dtype=ms.float16)

image = pipe(
    prompt="a photo of a cat holding a sign that says hello world",
    negative_prompt="",
    num_inference_steps=28,
    height=1024,
    width=1024,
    guidance_scale=7.0,
)[0][0]

image.save("sd3_hello_world.png")

Note: Stable Diffusion 3.5 can also be run using the SD3 pipeline, and all mentioned optimizations and techniques apply to it as well. In total there are three official models in the SD3 family: - stabilityai/stable-diffusion-3-medium-diffusers - stabilityai/stable-diffusion-3.5-large - stabilityai/stable-diffusion-3.5-large-turbo

Dropping the T5 Text Encoder during Inference

Removing the memory-intensive 4.7B parameter T5-XXL text encoder during inference can significantly decrease the memory requirements for SD3 with only a slight loss in performance.

import mindspore as ms
from mindone.diffusers import StableDiffusion3Pipeline

pipe = StableDiffusion3Pipeline.from_pretrained(
    "stabilityai/stable-diffusion-3-medium-diffusers",
    text_encoder_3=None,
    tokenizer_3=None,
    mindspore_dtype=ms.float16
)

image = pipe(
    prompt="a photo of a cat holding a sign that says hello world",
    negative_prompt="",
    num_inference_steps=28,
    height=1024,
    width=1024,
    guidance_scale=7.0,
)[0][0]

image.save("sd3_hello_world-no-T5.png")

Using Long Prompts with the T5 Text Encoder

By default, the T5 Text Encoder prompt uses a maximum sequence length of 256. This can be adjusted by setting the max_sequence_length to accept fewer or more tokens. Keep in mind that longer sequences require additional resources and result in longer generation times, such as during batch inference.

prompt = "A whimsical and creative image depicting a hybrid creature that is a mix of a waffle and a hippopotamus, basking in a river of melted butter amidst a breakfast-themed landscape. It features the distinctive, bulky body shape of a hippo. However, instead of the usual grey skin, the creature’s body resembles a golden-brown, crispy waffle fresh off the griddle. The skin is textured with the familiar grid pattern of a waffle, each square filled with a glistening sheen of syrup. The environment combines the natural habitat of a hippo with elements of a breakfast table setting, a river of warm, melted butter, with oversized utensils or plates peeking out from the lush, pancake-like foliage in the background, a towering pepper mill standing in for a tree.  As the sun rises in this fantastical world, it casts a warm, buttery glow over the scene. The creature, content in its butter river, lets out a yawn. Nearby, a flock of birds take flight"

image = pipe(
    prompt=prompt,
    negative_prompt="",
    num_inference_steps=28,
    guidance_scale=4.5,
    max_sequence_length=512,
)[0][0]

Sending a different prompt to the T5 Text Encoder

You can send a different prompt to the CLIP Text Encoders and the T5 Text Encoder to prevent the prompt from being truncated by the CLIP Text Encoders and to improve generation.

Tip

The prompt with the CLIP Text Encoders is still truncated to the 77 token limit.

prompt = "A whimsical and creative image depicting a hybrid creature that is a mix of a waffle and a hippopotamus, basking in a river of melted butter amidst a breakfast-themed landscape. A river of warm, melted butter, pancake-like foliage in the background, a towering pepper mill standing in for a tree."

prompt_3 = "A whimsical and creative image depicting a hybrid creature that is a mix of a waffle and a hippopotamus, basking in a river of melted butter amidst a breakfast-themed landscape. It features the distinctive, bulky body shape of a hippo. However, instead of the usual grey skin, the creature’s body resembles a golden-brown, crispy waffle fresh off the griddle. The skin is textured with the familiar grid pattern of a waffle, each square filled with a glistening sheen of syrup. The environment combines the natural habitat of a hippo with elements of a breakfast table setting, a river of warm, melted butter, with oversized utensils or plates peeking out from the lush, pancake-like foliage in the background, a towering pepper mill standing in for a tree.  As the sun rises in this fantastical world, it casts a warm, buttery glow over the scene. The creature, content in its butter river, lets out a yawn. Nearby, a flock of birds take flight"

image = pipe(
    prompt=prompt,
    prompt_3=prompt_3,
    negative_prompt="",
    num_inference_steps=28,
    guidance_scale=4.5,
    max_sequence_length=512,
)[0][0]

Tiny AutoEncoder for Stable Diffusion 3

Tiny AutoEncoder for Stable Diffusion (TAESD3) is a tiny distilled version of Stable Diffusion 3's VAE by Ollin Boer Bohan that can decode StableDiffusion3Pipeline latents almost instantly.

To use with Stable Diffusion 3:

import mindspore
from mindone.diffusers import StableDiffusion3Pipeline, AutoencoderTiny

pipe = StableDiffusion3Pipeline.from_pretrained(
    "stabilityai/stable-diffusion-3-medium-diffusers", mindspore_dtype=mindspore.float16
)
pipe.vae = AutoencoderTiny.from_pretrained("madebyollin/taesd3", mindspore_dtype=mindspore.float16)

prompt = "slice of delicious New York-style berry cheesecake"
image = pipe(prompt, num_inference_steps=25)[0][0]
image.save("cheesecake.png")

Loading the original checkpoints via from_single_file

The SD3Transformer2DModel and StableDiffusion3Pipeline classes support loading the original checkpoints via the from_single_file method. This method allows you to load the original checkpoint files that were used to train the models.

Loading the original checkpoints for the SD3Transformer2DModel

from mindone.diffusers import SD3Transformer2DModel

model = SD3Transformer2DModel.from_single_file("https://huggingface.co/stabilityai/stable-diffusion-3-medium/blob/main/sd3_medium.safetensors")

Loading the single checkpoint for the StableDiffusion3Pipeline

Loading the single file checkpoint without T5

import mindspore as ms
from mindone.diffusers import StableDiffusion3Pipeline

pipe = StableDiffusion3Pipeline.from_single_file(
    "https://huggingface.co/stabilityai/stable-diffusion-3-medium/blob/main/sd3_medium_incl_clips.safetensors",
    mindspore_dtype=ms.float16,
    text_encoder_3=None
)

image = pipe("a picture of a cat holding a sign that says hello world").images[0]
image.save('sd3-single-file.png')

Loading the single file checkpoint with T5

import mindspore as ms
from mindone.diffusers import StableDiffusion3Pipeline

pipe = StableDiffusion3Pipeline.from_single_file(
    "https://huggingface.co/stabilityai/stable-diffusion-3-medium/blob/main/sd3_medium_incl_clips_t5xxlfp8.safetensors",
    mindspore_dtype=ms.float16,
)

image = pipe("a picture of a cat holding a sign that says hello world")[0][0]
image.save('sd3-single-file-t5-fp8.png')

Loading the single file checkpoint for the Stable Diffusion 3.5 Transformer Model

import mindspore as ms
from mindone.diffusers import SD3Transformer2DModel, StableDiffusion3Pipeline

transformer = SD3Transformer2DModel.from_single_file(
    "https://huggingface.co/stabilityai/stable-diffusion-3.5-large-turbo/blob/main/sd3.5_large.safetensors",
    mindspore_dtype=ms.bfloat16,
)
pipe = StableDiffusion3Pipeline.from_pretrained(
    "stabilityai/stable-diffusion-3.5-large",
    transformer=transformer,
    mindspore_dtype=ms.bfloat16,
)

image = pipe("a cat holding a sign that says hello world")[0][0]
image.save("sd35.png")

mindone.diffusers.StableDiffusion3Pipeline

Bases: DiffusionPipeline, SD3LoraLoaderMixin, FromSingleFileMixin, SD3IPAdapterMixin

PARAMETER	DESCRIPTION
transformer	Conditional Transformer (MMDiT) architecture to denoise the encoded image latents. TYPE: [`SD3Transformer2DModel`]
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	CLIP, specifically the clip-vit-large-patch14 variant, with an additional added projection layer that is initialized with a diagonal matrix with the hidden_size as its dimension. TYPE: [`CLIPTextModelWithProjection`]
text_encoder_2	CLIP, specifically the laion/CLIP-ViT-bigG-14-laion2B-39B-b160k variant. TYPE: [`CLIPTextModelWithProjection`]
text_encoder_3	Frozen text-encoder. Stable Diffusion 3 uses T5, specifically the t5-v1_1-xxl variant. TYPE: [`T5EncoderModel`]
tokenizer	Tokenizer of class CLIPTokenizer. TYPE: `CLIPTokenizer`
tokenizer_2	Second Tokenizer of class CLIPTokenizer. TYPE: `CLIPTokenizer`
tokenizer_3	Tokenizer of class T5Tokenizer. TYPE: `T5TokenizerFast`
image_encoder	Pre-trained Vision Model for IP Adapter. TYPE: `MSPreTrainedModel`, *optional* DEFAULT: None
feature_extractor	Image processor for IP Adapter. TYPE: `BaseImageProcessor`, *optional* DEFAULT: None

Source code in mindone/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py

class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingleFileMixin, SD3IPAdapterMixin):
    r"""
    Args:
        transformer ([`SD3Transformer2DModel`]):
            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 ([`CLIPTextModelWithProjection`]):
            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection),
            specifically the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant,
            with an additional added projection layer that is initialized with a diagonal matrix with the `hidden_size`
            as its dimension.
        text_encoder_2 ([`CLIPTextModelWithProjection`]):
            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection),
            specifically the
            [laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)
            variant.
        text_encoder_3 ([`T5EncoderModel`]):
            Frozen text-encoder. Stable Diffusion 3 uses
            [T5](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5EncoderModel), specifically the
            [t5-v1_1-xxl](https://huggingface.co/google/t5-v1_1-xxl) variant.
        tokenizer (`CLIPTokenizer`):
            Tokenizer of class
            [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
        tokenizer_2 (`CLIPTokenizer`):
            Second Tokenizer of class
            [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
        tokenizer_3 (`T5TokenizerFast`):
            Tokenizer of class
            [T5Tokenizer](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5Tokenizer).
        image_encoder (`MSPreTrainedModel`, *optional*):
            Pre-trained Vision Model for IP Adapter.
        feature_extractor (`BaseImageProcessor`, *optional*):
            Image processor for IP Adapter.
    """

    model_cpu_offload_seq = "text_encoder->text_encoder_2->text_encoder_3->image_encoder->transformer->vae"
    _optional_components = ["image_encoder", "feature_extractor"]
    _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds", "negative_pooled_prompt_embeds"]

    def __init__(
        self,
        transformer: SD3Transformer2DModel,
        scheduler: FlowMatchEulerDiscreteScheduler,
        vae: AutoencoderKL,
        text_encoder: CLIPTextModelWithProjection,
        tokenizer: CLIPTokenizer,
        text_encoder_2: CLIPTextModelWithProjection,
        tokenizer_2: CLIPTokenizer,
        text_encoder_3: T5EncoderModel,
        tokenizer_3: T5TokenizerFast,
        image_encoder: MSPreTrainedModel = None,
        feature_extractor: BaseImageProcessor = None,
    ):
        super().__init__()

        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            text_encoder_2=text_encoder_2,
            text_encoder_3=text_encoder_3,
            tokenizer=tokenizer,
            tokenizer_2=tokenizer_2,
            tokenizer_3=tokenizer_3,
            transformer=transformer,
            scheduler=scheduler,
            image_encoder=image_encoder,
            feature_extractor=feature_extractor,
        )
        self.vae_scale_factor = (
            2 ** (len(self.vae.config.block_out_channels) - 1) if hasattr(self, "vae") and self.vae is not None else 8
        )
        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
        self.tokenizer_max_length = (
            self.tokenizer.model_max_length if hasattr(self, "tokenizer") and self.tokenizer is not None else 77
        )
        self.default_sample_size = (
            self.transformer.config.sample_size
            if hasattr(self, "transformer") and self.transformer is not None
            else 128
        )
        self.patch_size = (
            self.transformer.config.patch_size if hasattr(self, "transformer") and self.transformer is not None else 2
        )

        self.patch_size = (
            self.transformer.config.patch_size if hasattr(self, "transformer") and self.transformer is not None else 2
        )

    def _get_t5_prompt_embeds(
        self,
        prompt: Union[str, List[str]] = None,
        num_images_per_prompt: int = 1,
        max_sequence_length: int = 256,
        dtype=None,
    ):
        dtype = dtype or self.text_encoder.dtype

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

        if self.text_encoder_3 is None:
            return ops.zeros(
                (
                    batch_size * num_images_per_prompt,
                    self.tokenizer_max_length,
                    self.transformer.config.joint_attention_dim,
                ),
                dtype=dtype,
            )

        text_inputs = self.tokenizer_3(
            prompt,
            padding="max_length",
            max_length=max_sequence_length,
            truncation=True,
            add_special_tokens=True,
            return_tensors="np",
        )
        text_input_ids = text_inputs.input_ids
        untruncated_ids = self.tokenizer_3(prompt, padding="longest", return_tensors="np").input_ids

        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not np.array_equal(
            text_input_ids, untruncated_ids
        ):
            removed_text = self.tokenizer_3.batch_decode(untruncated_ids[:, self.tokenizer_max_length - 1 : -1])
            logger.warning(
                "The following part of your input was truncated because `max_sequence_length` is set to "
                f" {max_sequence_length} tokens: {removed_text}"
            )

        prompt_embeds = self.text_encoder_3(ms.Tensor.from_numpy(text_input_ids))[0]

        dtype = self.text_encoder_3.dtype
        prompt_embeds = prompt_embeds.to(dtype=dtype)

        _, seq_len, _ = prompt_embeds.shape

        # duplicate text embeddings and attention mask for each generation per prompt, using mps friendly method
        prompt_embeds = prompt_embeds.tile((1, num_images_per_prompt, 1))
        prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)

        return prompt_embeds

    def _get_clip_prompt_embeds(
        self,
        prompt: Union[str, List[str]],
        num_images_per_prompt: int = 1,
        clip_skip: Optional[int] = None,
        clip_model_index: int = 0,
    ):
        clip_tokenizers = [self.tokenizer, self.tokenizer_2]
        clip_text_encoders = [self.text_encoder, self.text_encoder_2]

        tokenizer = clip_tokenizers[clip_model_index]
        text_encoder = clip_text_encoders[clip_model_index]

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

        text_inputs = tokenizer(
            prompt,
            padding="max_length",
            max_length=self.tokenizer_max_length,
            truncation=True,
            return_tensors="np",
        )

        text_input_ids = text_inputs.input_ids
        untruncated_ids = tokenizer(prompt, padding="longest", return_tensors="np").input_ids
        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not np.array_equal(
            text_input_ids, untruncated_ids
        ):
            removed_text = tokenizer.batch_decode(untruncated_ids[:, self.tokenizer_max_length - 1 : -1])
            logger.warning(
                "The following part of your input was truncated because CLIP can only handle sequences up to"
                f" {self.tokenizer_max_length} tokens: {removed_text}"
            )
        prompt_embeds = text_encoder(ms.Tensor.from_numpy(text_input_ids), output_hidden_states=True)
        pooled_prompt_embeds = prompt_embeds[0]

        if clip_skip is None:
            prompt_embeds = prompt_embeds[2][-2]
        else:
            prompt_embeds = prompt_embeds[2][-(clip_skip + 2)]

        prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype)

        _, seq_len, _ = prompt_embeds.shape
        # duplicate text embeddings for each generation per prompt, using mps friendly method
        prompt_embeds = prompt_embeds.tile((1, num_images_per_prompt, 1))
        prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)

        pooled_prompt_embeds = pooled_prompt_embeds.tile((1, num_images_per_prompt, 1))
        pooled_prompt_embeds = pooled_prompt_embeds.view(batch_size * num_images_per_prompt, -1)

        return prompt_embeds, pooled_prompt_embeds

    def encode_prompt(
        self,
        prompt: Union[str, List[str]],
        prompt_2: Union[str, List[str]],
        prompt_3: Union[str, List[str]],
        num_images_per_prompt: int = 1,
        do_classifier_free_guidance: bool = True,
        negative_prompt: Optional[Union[str, List[str]]] = None,
        negative_prompt_2: Optional[Union[str, List[str]]] = None,
        negative_prompt_3: Optional[Union[str, List[str]]] = None,
        prompt_embeds: Optional[ms.Tensor] = None,
        negative_prompt_embeds: Optional[ms.Tensor] = None,
        pooled_prompt_embeds: Optional[ms.Tensor] = None,
        negative_pooled_prompt_embeds: Optional[ms.Tensor] = None,
        clip_skip: Optional[int] = None,
        max_sequence_length: int = 256,
        lora_scale: Optional[float] = None,
    ):
        r"""

        Args:
            prompt (`str` or `List[str]`, *optional*):
                prompt to be encoded
            prompt_2 (`str` or `List[str]`, *optional*):
                The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
                used in all text-encoders
            prompt_3 (`str` or `List[str]`, *optional*):
                The prompt or prompts to be sent to the `tokenizer_3` and `text_encoder_3`. If not defined, `prompt` is
                used in all text-encoders
            num_images_per_prompt (`int`):
                number of images that should be generated per prompt
            do_classifier_free_guidance (`bool`):
                whether to use classifier free guidance or not
            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 `guidance_scale` is
                less than `1`).
            negative_prompt_2 (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
                `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
            negative_prompt_2 (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation to be sent to `tokenizer_3` and
                `text_encoder_3`. If not defined, `negative_prompt` is used in both text-encoders
            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.
            pooled_prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
                If not provided, pooled text embeddings will be generated from `prompt` input argument.
            negative_pooled_prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
                input argument.
            clip_skip (`int`, *optional*):
                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                the output of the pre-final layer will be used for computing the prompt embeddings.
            lora_scale (`float`, *optional*):
                A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
        """
        # set lora scale so that monkey patched LoRA
        # function of text encoder can correctly access it
        if lora_scale is not None and isinstance(self, SD3LoraLoaderMixin):
            self._lora_scale = lora_scale

            # dynamically adjust the LoRA scale
            if self.text_encoder is not None:
                scale_lora_layers(self.text_encoder, lora_scale)
            if self.text_encoder_2 is not None:
                scale_lora_layers(self.text_encoder_2, lora_scale)

        prompt = [prompt] if isinstance(prompt, str) else prompt
        if prompt is not None:
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        if prompt_embeds is None:
            prompt_2 = prompt_2 or prompt
            prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2

            prompt_3 = prompt_3 or prompt
            prompt_3 = [prompt_3] if isinstance(prompt_3, str) else prompt_3

            prompt_embed, pooled_prompt_embed = self._get_clip_prompt_embeds(
                prompt=prompt,
                num_images_per_prompt=num_images_per_prompt,
                clip_skip=clip_skip,
                clip_model_index=0,
            )
            prompt_2_embed, pooled_prompt_2_embed = self._get_clip_prompt_embeds(
                prompt=prompt_2,
                num_images_per_prompt=num_images_per_prompt,
                clip_skip=clip_skip,
                clip_model_index=1,
            )
            clip_prompt_embeds = ops.cat([prompt_embed, prompt_2_embed], axis=-1)

            t5_prompt_embed = self._get_t5_prompt_embeds(
                prompt=prompt_3,
                num_images_per_prompt=num_images_per_prompt,
                max_sequence_length=max_sequence_length,
            )

            clip_prompt_embeds = pad(clip_prompt_embeds, (0, t5_prompt_embed.shape[-1] - clip_prompt_embeds.shape[-1]))

            prompt_embeds = ops.cat([clip_prompt_embeds, t5_prompt_embed], axis=-2)
            pooled_prompt_embeds = ops.cat([pooled_prompt_embed, pooled_prompt_2_embed], axis=-1)

        if do_classifier_free_guidance and negative_prompt_embeds is None:
            negative_prompt = negative_prompt or ""
            negative_prompt_2 = negative_prompt_2 or negative_prompt
            negative_prompt_3 = negative_prompt_3 or negative_prompt

            # normalize str to list
            negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
            negative_prompt_2 = (
                batch_size * [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2
            )
            negative_prompt_3 = (
                batch_size * [negative_prompt_3] if isinstance(negative_prompt_3, str) else negative_prompt_3
            )

            if prompt is not None and type(prompt) is not type(negative_prompt):
                raise TypeError(
                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
                    f" {type(prompt)}."
                )
            elif batch_size != len(negative_prompt):
                raise ValueError(
                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
                    " the batch size of `prompt`."
                )

            negative_prompt_embed, negative_pooled_prompt_embed = self._get_clip_prompt_embeds(
                negative_prompt,
                num_images_per_prompt=num_images_per_prompt,
                clip_skip=None,
                clip_model_index=0,
            )
            negative_prompt_2_embed, negative_pooled_prompt_2_embed = self._get_clip_prompt_embeds(
                negative_prompt_2,
                num_images_per_prompt=num_images_per_prompt,
                clip_skip=None,
                clip_model_index=1,
            )
            negative_clip_prompt_embeds = ops.cat([negative_prompt_embed, negative_prompt_2_embed], axis=-1)

            t5_negative_prompt_embed = self._get_t5_prompt_embeds(
                prompt=negative_prompt_3,
                num_images_per_prompt=num_images_per_prompt,
                max_sequence_length=max_sequence_length,
            )

            negative_clip_prompt_embeds = pad(
                negative_clip_prompt_embeds,
                (0, t5_negative_prompt_embed.shape[-1] - negative_clip_prompt_embeds.shape[-1]),
            )

            negative_prompt_embeds = ops.cat([negative_clip_prompt_embeds, t5_negative_prompt_embed], axis=-2)
            negative_pooled_prompt_embeds = ops.cat(
                [negative_pooled_prompt_embed, negative_pooled_prompt_2_embed], axis=-1
            )

        if self.text_encoder is not None:
            if isinstance(self, SD3LoraLoaderMixin):
                # Retrieve the original scale by scaling back the LoRA layers
                unscale_lora_layers(self.text_encoder, lora_scale)

        if self.text_encoder_2 is not None:
            if isinstance(self, SD3LoraLoaderMixin):
                # Retrieve the original scale by scaling back the LoRA layers
                unscale_lora_layers(self.text_encoder_2, lora_scale)

        return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds

    def check_inputs(
        self,
        prompt,
        prompt_2,
        prompt_3,
        height,
        width,
        negative_prompt=None,
        negative_prompt_2=None,
        negative_prompt_3=None,
        prompt_embeds=None,
        negative_prompt_embeds=None,
        pooled_prompt_embeds=None,
        negative_pooled_prompt_embeds=None,
        callback_on_step_end_tensor_inputs=None,
        max_sequence_length=None,
    ):
        if (
            height % (self.vae_scale_factor * self.patch_size) != 0
            or width % (self.vae_scale_factor * self.patch_size) != 0
        ):
            raise ValueError(
                f"`height` and `width` have to be divisible by {self.vae_scale_factor * self.patch_size} but are {height} and {width}."
                f"You can use height {height - height % (self.vae_scale_factor * self.patch_size)} and width {width - width % (self.vae_scale_factor * self.patch_size)}."  # noqa: E501
            )

        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 {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"  # noqa: E501
            )

        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_2 is not None and prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt_3 is not None and prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt_3`: {prompt_2} 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)}")
        elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
            raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
        elif prompt_3 is not None and (not isinstance(prompt_3, str) and not isinstance(prompt_3, list)):
            raise ValueError(f"`prompt_3` has to be of type `str` or `list` but is {type(prompt_3)}")

        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."
            )
        elif negative_prompt_2 is not None and negative_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} and `negative_prompt_embeds`:"
                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
            )
        elif negative_prompt_3 is not None and negative_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt_3`: {negative_prompt_3} 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 negative_prompt_embeds is not None:
            if prompt_embeds.shape != negative_prompt_embeds.shape:
                raise ValueError(
                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
                    f" {negative_prompt_embeds.shape}."
                )

        if prompt_embeds is not None and pooled_prompt_embeds is None:
            raise ValueError(
                "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`."  # noqa: E501
            )

        if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None:
            raise ValueError(
                "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."  # noqa: E501
            )

        if max_sequence_length is not None and max_sequence_length > 512:
            raise ValueError(f"`max_sequence_length` cannot be greater than 512 but is {max_sequence_length}")

    def prepare_latents(
        self,
        batch_size,
        num_channels_latents,
        height,
        width,
        dtype,
        generator,
        latents=None,
    ):
        if latents is not None:
            return latents.to(dtype=dtype)

        shape = (
            batch_size,
            num_channels_latents,
            int(height) // self.vae_scale_factor,
            int(width) // self.vae_scale_factor,
        )

        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)

        return latents

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

    @property
    def skip_guidance_layers(self):
        return self._skip_guidance_layers

    @property
    def clip_skip(self):
        return self._clip_skip

    # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
    # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
    # corresponds to doing no classifier free guidance.
    @property
    def do_classifier_free_guidance(self):
        return self._guidance_scale > 1

    @property
    def joint_attention_kwargs(self):
        return self._joint_attention_kwargs

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

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

    # Adapted from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.encode_image
    def encode_image(self, image: PipelineImageInput) -> ms.Tensor:
        """Encodes the given image into a feature representation using a pre-trained image encoder.

        Args:
            image (`PipelineImageInput`):
                Input image to be encoded.

        Returns:
            `ms.Tensor`: The encoded image feature representation.
        """
        if not isinstance(image, ms.Tensor):
            image = self.feature_extractor(image, return_tensors="np").pixel_values
            image = ms.tensor(image)

        image = image.to(dtype=self.dtype)

        return self.image_encoder(image, output_hidden_states=True)[2][-2]

    # Adapted from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.prepare_ip_adapter_image_embeds
    def prepare_ip_adapter_image_embeds(
        self,
        ip_adapter_image: Optional[PipelineImageInput] = None,
        ip_adapter_image_embeds: Optional[ms.Tensor] = None,
        num_images_per_prompt: int = 1,
        do_classifier_free_guidance: bool = True,
    ) -> ms.Tensor:
        """Prepares image embeddings for use in the IP-Adapter.

        Either `ip_adapter_image` or `ip_adapter_image_embeds` must be passed.

        Args:
            ip_adapter_image (`PipelineImageInput`, *optional*):
                The input image to extract features from for IP-Adapter.
            ip_adapter_image_embeds (`ms.Tensor`, *optional*):
                Precomputed image embeddings.
            num_images_per_prompt (`int`, defaults to 1):
                Number of images that should be generated per prompt.
            do_classifier_free_guidance (`bool`, defaults to True):
                Whether to use classifier free guidance or not.
        """
        if ip_adapter_image_embeds is not None:
            if do_classifier_free_guidance:
                single_negative_image_embeds, single_image_embeds = ip_adapter_image_embeds.chunk(2)
            else:
                single_image_embeds = ip_adapter_image_embeds
        elif ip_adapter_image is not None:
            single_image_embeds = self.encode_image(ip_adapter_image)
            if do_classifier_free_guidance:
                single_negative_image_embeds = ops.zeros_like(single_image_embeds)
        else:
            raise ValueError("Neither `ip_adapter_image_embeds` or `ip_adapter_image_embeds` were provided.")

        image_embeds = ops.cat([single_image_embeds] * num_images_per_prompt, axis=0)

        if do_classifier_free_guidance:
            negative_image_embeds = ops.cat([single_negative_image_embeds] * num_images_per_prompt, axis=0)
            image_embeds = ops.cat([negative_image_embeds, image_embeds], axis=0)

        return image_embeds

    def __call__(
        self,
        prompt: Union[str, List[str]] = None,
        prompt_2: Optional[Union[str, List[str]]] = None,
        prompt_3: Optional[Union[str, List[str]]] = None,
        height: Optional[int] = None,
        width: Optional[int] = None,
        num_inference_steps: int = 28,
        sigmas: Optional[List[float]] = None,
        guidance_scale: float = 7.0,
        negative_prompt: Optional[Union[str, List[str]]] = None,
        negative_prompt_2: Optional[Union[str, List[str]]] = None,
        negative_prompt_3: Optional[Union[str, List[str]]] = None,
        num_images_per_prompt: Optional[int] = 1,
        generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
        latents: Optional[ms.Tensor] = None,
        prompt_embeds: Optional[ms.Tensor] = None,
        negative_prompt_embeds: Optional[ms.Tensor] = None,
        pooled_prompt_embeds: Optional[ms.Tensor] = None,
        negative_pooled_prompt_embeds: Optional[ms.Tensor] = None,
        ip_adapter_image: Optional[PipelineImageInput] = None,
        ip_adapter_image_embeds: Optional[ms.Tensor] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = False,
        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
        clip_skip: Optional[int] = 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 = 256,
        skip_guidance_layers: List[int] = None,
        skip_layer_guidance_scale: float = 2.8,
        skip_layer_guidance_stop: float = 0.2,
        skip_layer_guidance_start: float = 0.01,
        mu: Optional[float] = None,
    ):
        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.
            prompt_2 (`str` or `List[str]`, *optional*):
                The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
                will be used instead
            prompt_3 (`str` or `List[str]`, *optional*):
                The prompt or prompts to be sent to `tokenizer_3` and `text_encoder_3`. If not defined, `prompt` is
                will be used instead
            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 7.0):
                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                `guidance_scale` is defined as `w` of equation 2. of [Imagen
                Paper](https://arxiv.org/pdf/2205.11487.pdf). 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.
            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 `guidance_scale` is
                less than `1`).
            negative_prompt_2 (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
                `text_encoder_2`. If not defined, `negative_prompt` is used instead
            negative_prompt_3 (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation to be sent to `tokenizer_3` and
                `text_encoder_3`. If not defined, `negative_prompt` is used instead
            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://pytorch.org/docs/stable/generated/torch.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 ge 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.
            pooled_prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
                If not provided, pooled text embeddings will be generated from `prompt` input argument.
            negative_pooled_prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
                input argument.
            ip_adapter_image (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
            ip_adapter_image_embeds (`ms.Tensor`, *optional*):
                Pre-generated image embeddings for IP-Adapter. Should be a tensor of shape `(batch_size, num_images,
                emb_dim)`. It should contain the negative image embedding if `do_classifier_free_guidance` is set to
                `True`. If not provided, embeddings are computed from the `ip_adapter_image` 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 `False`):
                Whether or not to return a [`~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput`] instead of
                a plain tuple.
            joint_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 256): Maximum sequence length to use with the `prompt`.
            skip_guidance_layers (`List[int]`, *optional*):
                A list of integers that specify layers to skip during guidance. If not provided, all layers will be
                used for guidance. If provided, the guidance will only be applied to the layers specified in the list.
                Recommended value by StabiltyAI for Stable Diffusion 3.5 Medium is [7, 8, 9].
            skip_layer_guidance_scale (`int`, *optional*): The scale of the guidance for the layers specified in
                `skip_guidance_layers`. The guidance will be applied to the layers specified in `skip_guidance_layers`
                with a scale of `skip_layer_guidance_scale`. The guidance will be applied to the rest of the layers
                with a scale of `1`.
            skip_layer_guidance_stop (`int`, *optional*): The step at which the guidance for the layers specified in
                `skip_guidance_layers` will stop. The guidance will be applied to the layers specified in
                `skip_guidance_layers` until the fraction specified in `skip_layer_guidance_stop`. Recommended value by
                StabiltyAI for Stable Diffusion 3.5 Medium is 0.2.
            skip_layer_guidance_start (`int`, *optional*): The step at which the guidance for the layers specified in
                `skip_guidance_layers` will start. The guidance will be applied to the layers specified in
                `skip_guidance_layers` from the fraction specified in `skip_layer_guidance_start`. Recommended value by
                StabiltyAI for Stable Diffusion 3.5 Medium is 0.01.
            mu (`float`, *optional*): `mu` value used for `dynamic_shifting`.

        Examples:

        Returns:
            [`~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput`] or `tuple`:
            [`~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput`] 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,
            prompt_2,
            prompt_3,
            height,
            width,
            negative_prompt=negative_prompt,
            negative_prompt_2=negative_prompt_2,
            negative_prompt_3=negative_prompt_3,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            pooled_prompt_embeds=pooled_prompt_embeds,
            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
            callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
            max_sequence_length=max_sequence_length,
        )

        self._guidance_scale = guidance_scale
        self._skip_layer_guidance_scale = skip_layer_guidance_scale
        self._clip_skip = clip_skip
        self._joint_attention_kwargs = joint_attention_kwargs
        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]

        # we're popping the `scale` instead of getting it because otherwise `scale` will be propagated
        # to the transformer and will raise RuntimeError.
        lora_scale = self.joint_attention_kwargs.pop("scale", None) if self.joint_attention_kwargs is not None else None

        (
            prompt_embeds,
            negative_prompt_embeds,
            pooled_prompt_embeds,
            negative_pooled_prompt_embeds,
        ) = self.encode_prompt(
            prompt=prompt,
            prompt_2=prompt_2,
            prompt_3=prompt_3,
            negative_prompt=negative_prompt,
            negative_prompt_2=negative_prompt_2,
            negative_prompt_3=negative_prompt_3,
            do_classifier_free_guidance=self.do_classifier_free_guidance,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            pooled_prompt_embeds=pooled_prompt_embeds,
            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
            clip_skip=self.clip_skip,
            num_images_per_prompt=num_images_per_prompt,
            max_sequence_length=max_sequence_length,
            lora_scale=lora_scale,
        )

        if self.do_classifier_free_guidance:
            if skip_guidance_layers is not None:
                original_prompt_embeds = prompt_embeds
                original_pooled_prompt_embeds = pooled_prompt_embeds
            prompt_embeds = ops.cat([negative_prompt_embeds, prompt_embeds], axis=0)
            pooled_prompt_embeds = ops.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds], axis=0)

        # 4. Prepare latent variables
        num_channels_latents = self.transformer.config.in_channels
        latents = self.prepare_latents(
            batch_size * num_images_per_prompt,
            num_channels_latents,
            height,
            width,
            prompt_embeds.dtype,
            generator,
            latents,
        )

        # 5. Prepare timesteps
        scheduler_kwargs = {}
        if self.scheduler.config.get("use_dynamic_shifting", None) and mu is None:
            _, _, height, width = latents.shape
            image_seq_len = (height // self.transformer.config.patch_size) * (
                width // self.transformer.config.patch_size
            )
            mu = calculate_shift(
                image_seq_len,
                self.scheduler.config.base_image_seq_len,
                self.scheduler.config.max_image_seq_len,
                self.scheduler.config.base_shift,
                self.scheduler.config.max_shift,
            )
            scheduler_kwargs["mu"] = mu
        elif mu is not None:
            scheduler_kwargs["mu"] = mu
        timesteps, num_inference_steps = retrieve_timesteps(
            self.scheduler,
            num_inference_steps,
            sigmas=sigmas,
            **scheduler_kwargs,
        )
        num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
        self._num_timesteps = len(timesteps)

        # 6. Prepare image embeddings
        if (ip_adapter_image is not None and self.is_ip_adapter_active) or ip_adapter_image_embeds is not None:
            ip_adapter_image_embeds = self.prepare_ip_adapter_image_embeds(
                ip_adapter_image,
                ip_adapter_image_embeds,
                batch_size * num_images_per_prompt,
                self.do_classifier_free_guidance,
            )

            if self.joint_attention_kwargs is None:
                self._joint_attention_kwargs = {"ip_adapter_image_embeds": ip_adapter_image_embeds}
            else:
                self._joint_attention_kwargs.update(ip_adapter_image_embeds=ip_adapter_image_embeds)

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

                # expand the latents if we are doing classifier free guidance
                latent_model_input = ops.cat([latents] * 2) if self.do_classifier_free_guidance else latents
                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
                timestep = t.broadcast_to((latent_model_input.shape[0],))

                noise_pred = self.transformer(
                    hidden_states=latent_model_input,
                    timestep=timestep,
                    encoder_hidden_states=prompt_embeds,
                    pooled_projections=pooled_prompt_embeds,
                    joint_attention_kwargs=self.joint_attention_kwargs,
                    return_dict=False,
                )[0]

                # perform guidance
                if self.do_classifier_free_guidance:
                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                    noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond)
                    should_skip_layers = (
                        True
                        if i > num_inference_steps * skip_layer_guidance_start
                        and i < num_inference_steps * skip_layer_guidance_stop
                        else False
                    )
                    if skip_guidance_layers is not None and should_skip_layers:
                        timestep = t.broadcast_to((latents.shape[0],))
                        latent_model_input = latents
                        noise_pred_skip_layers = self.transformer(
                            hidden_states=latent_model_input,
                            timestep=timestep,
                            encoder_hidden_states=original_prompt_embeds,
                            pooled_projections=original_pooled_prompt_embeds,
                            joint_attention_kwargs=self.joint_attention_kwargs,
                            return_dict=False,
                            skip_layers=skip_guidance_layers,
                        )[0]
                        noise_pred = (
                            noise_pred + (noise_pred_text - noise_pred_skip_layers) * self._skip_layer_guidance_scale
                        )

                # 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)
                    negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
                    negative_pooled_prompt_embeds = callback_outputs.pop(
                        "negative_pooled_prompt_embeds", negative_pooled_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()

        if lora_scale is not None:
            # remove `lora_scale` from each PEFT layer
            unscale_lora_layers(self.transformer, lora_scale)

        if output_type == "latent":
            image = latents

        else:
            latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor
            latents = latents.to(
                self.vae.dtype
            )  # for validation in training where vae and transformer might have different dtype

            image = self.vae.decode(latents, return_dict=False)[0]
            image = self.image_processor.postprocess(image, output_type=output_type)

        if not return_dict:
            return (image,)

        return StableDiffusion3PipelineOutput(images=image)

mindone.diffusers.StableDiffusion3Pipeline.__call__(prompt=None, prompt_2=None, prompt_3=None, height=None, width=None, num_inference_steps=28, sigmas=None, guidance_scale=7.0, negative_prompt=None, negative_prompt_2=None, negative_prompt_3=None, num_images_per_prompt=1, generator=None, latents=None, prompt_embeds=None, negative_prompt_embeds=None, pooled_prompt_embeds=None, negative_pooled_prompt_embeds=None, ip_adapter_image=None, ip_adapter_image_embeds=None, output_type='pil', return_dict=False, joint_attention_kwargs=None, clip_skip=None, callback_on_step_end=None, callback_on_step_end_tensor_inputs=['latents'], max_sequence_length=256, skip_guidance_layers=None, skip_layer_guidance_scale=2.8, skip_layer_guidance_stop=0.2, skip_layer_guidance_start=0.01, mu=None)

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
prompt_2	The prompt or prompts to be sent to tokenizer_2 and text_encoder_2. If not defined, prompt is will be used instead TYPE: `str` or `List[str]`, *optional* DEFAULT: None
prompt_3	The prompt or prompts to be sent to tokenizer_3 and text_encoder_3. If not defined, prompt is will be used instead TYPE: `str` or `List[str]`, *optional* DEFAULT: None
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: 28
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	Guidance scale as defined in Classifier-Free Diffusion Guidance. guidance_scale is defined as w of equation 2. of Imagen Paper. 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. TYPE: `float`, *optional*, defaults to 7.0 DEFAULT: 7.0
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 guidance_scale is less than 1). TYPE: `str` or `List[str]`, *optional* DEFAULT: None
negative_prompt_2	The prompt or prompts not to guide the image generation to be sent to tokenizer_2 and text_encoder_2. If not defined, negative_prompt is used instead TYPE: `str` or `List[str]`, *optional* DEFAULT: None
negative_prompt_3	The prompt or prompts not to guide the image generation to be sent to tokenizer_3 and text_encoder_3. If not defined, negative_prompt is used instead TYPE: `str` or `List[str]`, *optional* 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 ge 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
pooled_prompt_embeds	Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, pooled text embeddings will be generated from prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
negative_pooled_prompt_embeds	Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, pooled negative_prompt_embeds will be generated from negative_prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
ip_adapter_image	Optional image input to work with IP Adapters. TYPE: `PipelineImageInput`, *optional* DEFAULT: None
ip_adapter_image_embeds	Pre-generated image embeddings for IP-Adapter. Should be a tensor of shape (batch_size, num_images, emb_dim). It should contain the negative image embedding if do_classifier_free_guidance is set to True. If not provided, embeddings are computed from the ip_adapter_image 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.stable_diffusion_3.StableDiffusion3PipelineOutput] instead of a plain tuple. TYPE: `bool`, *optional*, defaults to `False` DEFAULT: False
joint_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 256 DEFAULT: 256
skip_guidance_layers	A list of integers that specify layers to skip during guidance. If not provided, all layers will be used for guidance. If provided, the guidance will only be applied to the layers specified in the list. Recommended value by StabiltyAI for Stable Diffusion 3.5 Medium is [7, 8, 9]. TYPE: `List[int]`, *optional* DEFAULT: None
skip_layer_guidance_scale	The scale of the guidance for the layers specified in skip_guidance_layers. The guidance will be applied to the layers specified in skip_guidance_layers with a scale of skip_layer_guidance_scale. The guidance will be applied to the rest of the layers with a scale of 1. TYPE: `int`, *optional* DEFAULT: 2.8
skip_layer_guidance_stop	The step at which the guidance for the layers specified in skip_guidance_layers will stop. The guidance will be applied to the layers specified in skip_guidance_layers until the fraction specified in skip_layer_guidance_stop. Recommended value by StabiltyAI for Stable Diffusion 3.5 Medium is 0.2. TYPE: `int`, *optional* DEFAULT: 0.2
skip_layer_guidance_start	The step at which the guidance for the layers specified in skip_guidance_layers will start. The guidance will be applied to the layers specified in skip_guidance_layers from the fraction specified in skip_layer_guidance_start. Recommended value by StabiltyAI for Stable Diffusion 3.5 Medium is 0.01. TYPE: `int`, *optional* DEFAULT: 0.01
mu	mu value used for dynamic_shifting. TYPE: `float`, *optional* DEFAULT: None

RETURNS	DESCRIPTION
	[~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput] or tuple:
	[~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput] 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/stable_diffusion_3/pipeline_stable_diffusion_3.py

def __call__(
    self,
    prompt: Union[str, List[str]] = None,
    prompt_2: Optional[Union[str, List[str]]] = None,
    prompt_3: Optional[Union[str, List[str]]] = None,
    height: Optional[int] = None,
    width: Optional[int] = None,
    num_inference_steps: int = 28,
    sigmas: Optional[List[float]] = None,
    guidance_scale: float = 7.0,
    negative_prompt: Optional[Union[str, List[str]]] = None,
    negative_prompt_2: Optional[Union[str, List[str]]] = None,
    negative_prompt_3: Optional[Union[str, List[str]]] = None,
    num_images_per_prompt: Optional[int] = 1,
    generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
    latents: Optional[ms.Tensor] = None,
    prompt_embeds: Optional[ms.Tensor] = None,
    negative_prompt_embeds: Optional[ms.Tensor] = None,
    pooled_prompt_embeds: Optional[ms.Tensor] = None,
    negative_pooled_prompt_embeds: Optional[ms.Tensor] = None,
    ip_adapter_image: Optional[PipelineImageInput] = None,
    ip_adapter_image_embeds: Optional[ms.Tensor] = None,
    output_type: Optional[str] = "pil",
    return_dict: bool = False,
    joint_attention_kwargs: Optional[Dict[str, Any]] = None,
    clip_skip: Optional[int] = 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 = 256,
    skip_guidance_layers: List[int] = None,
    skip_layer_guidance_scale: float = 2.8,
    skip_layer_guidance_stop: float = 0.2,
    skip_layer_guidance_start: float = 0.01,
    mu: Optional[float] = None,
):
    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.
        prompt_2 (`str` or `List[str]`, *optional*):
            The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
            will be used instead
        prompt_3 (`str` or `List[str]`, *optional*):
            The prompt or prompts to be sent to `tokenizer_3` and `text_encoder_3`. If not defined, `prompt` is
            will be used instead
        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 7.0):
            Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
            `guidance_scale` is defined as `w` of equation 2. of [Imagen
            Paper](https://arxiv.org/pdf/2205.11487.pdf). 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.
        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 `guidance_scale` is
            less than `1`).
        negative_prompt_2 (`str` or `List[str]`, *optional*):
            The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
            `text_encoder_2`. If not defined, `negative_prompt` is used instead
        negative_prompt_3 (`str` or `List[str]`, *optional*):
            The prompt or prompts not to guide the image generation to be sent to `tokenizer_3` and
            `text_encoder_3`. If not defined, `negative_prompt` is used instead
        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://pytorch.org/docs/stable/generated/torch.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 ge 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.
        pooled_prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
            If not provided, pooled text embeddings will be generated from `prompt` input argument.
        negative_pooled_prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
            weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
            input argument.
        ip_adapter_image (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
        ip_adapter_image_embeds (`ms.Tensor`, *optional*):
            Pre-generated image embeddings for IP-Adapter. Should be a tensor of shape `(batch_size, num_images,
            emb_dim)`. It should contain the negative image embedding if `do_classifier_free_guidance` is set to
            `True`. If not provided, embeddings are computed from the `ip_adapter_image` 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 `False`):
            Whether or not to return a [`~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput`] instead of
            a plain tuple.
        joint_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 256): Maximum sequence length to use with the `prompt`.
        skip_guidance_layers (`List[int]`, *optional*):
            A list of integers that specify layers to skip during guidance. If not provided, all layers will be
            used for guidance. If provided, the guidance will only be applied to the layers specified in the list.
            Recommended value by StabiltyAI for Stable Diffusion 3.5 Medium is [7, 8, 9].
        skip_layer_guidance_scale (`int`, *optional*): The scale of the guidance for the layers specified in
            `skip_guidance_layers`. The guidance will be applied to the layers specified in `skip_guidance_layers`
            with a scale of `skip_layer_guidance_scale`. The guidance will be applied to the rest of the layers
            with a scale of `1`.
        skip_layer_guidance_stop (`int`, *optional*): The step at which the guidance for the layers specified in
            `skip_guidance_layers` will stop. The guidance will be applied to the layers specified in
            `skip_guidance_layers` until the fraction specified in `skip_layer_guidance_stop`. Recommended value by
            StabiltyAI for Stable Diffusion 3.5 Medium is 0.2.
        skip_layer_guidance_start (`int`, *optional*): The step at which the guidance for the layers specified in
            `skip_guidance_layers` will start. The guidance will be applied to the layers specified in
            `skip_guidance_layers` from the fraction specified in `skip_layer_guidance_start`. Recommended value by
            StabiltyAI for Stable Diffusion 3.5 Medium is 0.01.
        mu (`float`, *optional*): `mu` value used for `dynamic_shifting`.

    Examples:

    Returns:
        [`~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput`] or `tuple`:
        [`~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput`] 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,
        prompt_2,
        prompt_3,
        height,
        width,
        negative_prompt=negative_prompt,
        negative_prompt_2=negative_prompt_2,
        negative_prompt_3=negative_prompt_3,
        prompt_embeds=prompt_embeds,
        negative_prompt_embeds=negative_prompt_embeds,
        pooled_prompt_embeds=pooled_prompt_embeds,
        negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
        callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
        max_sequence_length=max_sequence_length,
    )

    self._guidance_scale = guidance_scale
    self._skip_layer_guidance_scale = skip_layer_guidance_scale
    self._clip_skip = clip_skip
    self._joint_attention_kwargs = joint_attention_kwargs
    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]

    # we're popping the `scale` instead of getting it because otherwise `scale` will be propagated
    # to the transformer and will raise RuntimeError.
    lora_scale = self.joint_attention_kwargs.pop("scale", None) if self.joint_attention_kwargs is not None else None

    (
        prompt_embeds,
        negative_prompt_embeds,
        pooled_prompt_embeds,
        negative_pooled_prompt_embeds,
    ) = self.encode_prompt(
        prompt=prompt,
        prompt_2=prompt_2,
        prompt_3=prompt_3,
        negative_prompt=negative_prompt,
        negative_prompt_2=negative_prompt_2,
        negative_prompt_3=negative_prompt_3,
        do_classifier_free_guidance=self.do_classifier_free_guidance,
        prompt_embeds=prompt_embeds,
        negative_prompt_embeds=negative_prompt_embeds,
        pooled_prompt_embeds=pooled_prompt_embeds,
        negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
        clip_skip=self.clip_skip,
        num_images_per_prompt=num_images_per_prompt,
        max_sequence_length=max_sequence_length,
        lora_scale=lora_scale,
    )

    if self.do_classifier_free_guidance:
        if skip_guidance_layers is not None:
            original_prompt_embeds = prompt_embeds
            original_pooled_prompt_embeds = pooled_prompt_embeds
        prompt_embeds = ops.cat([negative_prompt_embeds, prompt_embeds], axis=0)
        pooled_prompt_embeds = ops.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds], axis=0)

    # 4. Prepare latent variables
    num_channels_latents = self.transformer.config.in_channels
    latents = self.prepare_latents(
        batch_size * num_images_per_prompt,
        num_channels_latents,
        height,
        width,
        prompt_embeds.dtype,
        generator,
        latents,
    )

    # 5. Prepare timesteps
    scheduler_kwargs = {}
    if self.scheduler.config.get("use_dynamic_shifting", None) and mu is None:
        _, _, height, width = latents.shape
        image_seq_len = (height // self.transformer.config.patch_size) * (
            width // self.transformer.config.patch_size
        )
        mu = calculate_shift(
            image_seq_len,
            self.scheduler.config.base_image_seq_len,
            self.scheduler.config.max_image_seq_len,
            self.scheduler.config.base_shift,
            self.scheduler.config.max_shift,
        )
        scheduler_kwargs["mu"] = mu
    elif mu is not None:
        scheduler_kwargs["mu"] = mu
    timesteps, num_inference_steps = retrieve_timesteps(
        self.scheduler,
        num_inference_steps,
        sigmas=sigmas,
        **scheduler_kwargs,
    )
    num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
    self._num_timesteps = len(timesteps)

    # 6. Prepare image embeddings
    if (ip_adapter_image is not None and self.is_ip_adapter_active) or ip_adapter_image_embeds is not None:
        ip_adapter_image_embeds = self.prepare_ip_adapter_image_embeds(
            ip_adapter_image,
            ip_adapter_image_embeds,
            batch_size * num_images_per_prompt,
            self.do_classifier_free_guidance,
        )

        if self.joint_attention_kwargs is None:
            self._joint_attention_kwargs = {"ip_adapter_image_embeds": ip_adapter_image_embeds}
        else:
            self._joint_attention_kwargs.update(ip_adapter_image_embeds=ip_adapter_image_embeds)

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

            # expand the latents if we are doing classifier free guidance
            latent_model_input = ops.cat([latents] * 2) if self.do_classifier_free_guidance else latents
            # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
            timestep = t.broadcast_to((latent_model_input.shape[0],))

            noise_pred = self.transformer(
                hidden_states=latent_model_input,
                timestep=timestep,
                encoder_hidden_states=prompt_embeds,
                pooled_projections=pooled_prompt_embeds,
                joint_attention_kwargs=self.joint_attention_kwargs,
                return_dict=False,
            )[0]

            # perform guidance
            if self.do_classifier_free_guidance:
                noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond)
                should_skip_layers = (
                    True
                    if i > num_inference_steps * skip_layer_guidance_start
                    and i < num_inference_steps * skip_layer_guidance_stop
                    else False
                )
                if skip_guidance_layers is not None and should_skip_layers:
                    timestep = t.broadcast_to((latents.shape[0],))
                    latent_model_input = latents
                    noise_pred_skip_layers = self.transformer(
                        hidden_states=latent_model_input,
                        timestep=timestep,
                        encoder_hidden_states=original_prompt_embeds,
                        pooled_projections=original_pooled_prompt_embeds,
                        joint_attention_kwargs=self.joint_attention_kwargs,
                        return_dict=False,
                        skip_layers=skip_guidance_layers,
                    )[0]
                    noise_pred = (
                        noise_pred + (noise_pred_text - noise_pred_skip_layers) * self._skip_layer_guidance_scale
                    )

            # 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)
                negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
                negative_pooled_prompt_embeds = callback_outputs.pop(
                    "negative_pooled_prompt_embeds", negative_pooled_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()

    if lora_scale is not None:
        # remove `lora_scale` from each PEFT layer
        unscale_lora_layers(self.transformer, lora_scale)

    if output_type == "latent":
        image = latents

    else:
        latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor
        latents = latents.to(
            self.vae.dtype
        )  # for validation in training where vae and transformer might have different dtype

        image = self.vae.decode(latents, return_dict=False)[0]
        image = self.image_processor.postprocess(image, output_type=output_type)

    if not return_dict:
        return (image,)

    return StableDiffusion3PipelineOutput(images=image)

mindone.diffusers.StableDiffusion3Pipeline.encode_image(image)

Encodes the given image into a feature representation using a pre-trained image encoder.

PARAMETER	DESCRIPTION
image	Input image to be encoded. TYPE: `PipelineImageInput`

RETURNS	DESCRIPTION
Tensor	ms.Tensor: The encoded image feature representation.

Source code in mindone/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py

def encode_image(self, image: PipelineImageInput) -> ms.Tensor:
    """Encodes the given image into a feature representation using a pre-trained image encoder.

    Args:
        image (`PipelineImageInput`):
            Input image to be encoded.

    Returns:
        `ms.Tensor`: The encoded image feature representation.
    """
    if not isinstance(image, ms.Tensor):
        image = self.feature_extractor(image, return_tensors="np").pixel_values
        image = ms.tensor(image)

    image = image.to(dtype=self.dtype)

    return self.image_encoder(image, output_hidden_states=True)[2][-2]

mindone.diffusers.StableDiffusion3Pipeline.encode_prompt(prompt, prompt_2, prompt_3, num_images_per_prompt=1, do_classifier_free_guidance=True, negative_prompt=None, negative_prompt_2=None, negative_prompt_3=None, prompt_embeds=None, negative_prompt_embeds=None, pooled_prompt_embeds=None, negative_pooled_prompt_embeds=None, clip_skip=None, max_sequence_length=256, lora_scale=None)

PARAMETER	DESCRIPTION
prompt	prompt to be encoded TYPE: `str` or `List[str]`, *optional*
prompt_2	The prompt or prompts to be sent to the tokenizer_2 and text_encoder_2. If not defined, prompt is used in all text-encoders TYPE: `str` or `List[str]`, *optional*
prompt_3	The prompt or prompts to be sent to the tokenizer_3 and text_encoder_3. If not defined, prompt is used in all text-encoders TYPE: `str` or `List[str]`, *optional*
num_images_per_prompt	number of images that should be generated per prompt TYPE: `int` DEFAULT: 1
do_classifier_free_guidance	whether to use classifier free guidance or not TYPE: `bool` DEFAULT: True
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 guidance_scale is less than 1). TYPE: `str` or `List[str]`, *optional* DEFAULT: None
negative_prompt_2	The prompt or prompts not to guide the image generation to be sent to tokenizer_2 and text_encoder_2. If not defined, negative_prompt is used in all the text-encoders. TYPE: `str` or `List[str]`, *optional* DEFAULT: None
negative_prompt_2	The prompt or prompts not to guide the image generation to be sent to tokenizer_3 and text_encoder_3. If not defined, negative_prompt is used in both text-encoders TYPE: `str` or `List[str]`, *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
pooled_prompt_embeds	Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, pooled text embeddings will be generated from prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
negative_pooled_prompt_embeds	Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, pooled negative_prompt_embeds will be generated from negative_prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
clip_skip	Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that the output of the pre-final layer will be used for computing the prompt embeddings. TYPE: `int`, *optional* DEFAULT: None
lora_scale	A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. TYPE: `float`, *optional* DEFAULT: None

Source code in mindone/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py

def encode_prompt(
    self,
    prompt: Union[str, List[str]],
    prompt_2: Union[str, List[str]],
    prompt_3: Union[str, List[str]],
    num_images_per_prompt: int = 1,
    do_classifier_free_guidance: bool = True,
    negative_prompt: Optional[Union[str, List[str]]] = None,
    negative_prompt_2: Optional[Union[str, List[str]]] = None,
    negative_prompt_3: Optional[Union[str, List[str]]] = None,
    prompt_embeds: Optional[ms.Tensor] = None,
    negative_prompt_embeds: Optional[ms.Tensor] = None,
    pooled_prompt_embeds: Optional[ms.Tensor] = None,
    negative_pooled_prompt_embeds: Optional[ms.Tensor] = None,
    clip_skip: Optional[int] = None,
    max_sequence_length: int = 256,
    lora_scale: Optional[float] = None,
):
    r"""

    Args:
        prompt (`str` or `List[str]`, *optional*):
            prompt to be encoded
        prompt_2 (`str` or `List[str]`, *optional*):
            The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
            used in all text-encoders
        prompt_3 (`str` or `List[str]`, *optional*):
            The prompt or prompts to be sent to the `tokenizer_3` and `text_encoder_3`. If not defined, `prompt` is
            used in all text-encoders
        num_images_per_prompt (`int`):
            number of images that should be generated per prompt
        do_classifier_free_guidance (`bool`):
            whether to use classifier free guidance or not
        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 `guidance_scale` is
            less than `1`).
        negative_prompt_2 (`str` or `List[str]`, *optional*):
            The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
            `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
        negative_prompt_2 (`str` or `List[str]`, *optional*):
            The prompt or prompts not to guide the image generation to be sent to `tokenizer_3` and
            `text_encoder_3`. If not defined, `negative_prompt` is used in both text-encoders
        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.
        pooled_prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
            If not provided, pooled text embeddings will be generated from `prompt` input argument.
        negative_pooled_prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
            weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
            input argument.
        clip_skip (`int`, *optional*):
            Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
            the output of the pre-final layer will be used for computing the prompt embeddings.
        lora_scale (`float`, *optional*):
            A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
    """
    # set lora scale so that monkey patched LoRA
    # function of text encoder can correctly access it
    if lora_scale is not None and isinstance(self, SD3LoraLoaderMixin):
        self._lora_scale = lora_scale

        # dynamically adjust the LoRA scale
        if self.text_encoder is not None:
            scale_lora_layers(self.text_encoder, lora_scale)
        if self.text_encoder_2 is not None:
            scale_lora_layers(self.text_encoder_2, lora_scale)

    prompt = [prompt] if isinstance(prompt, str) else prompt
    if prompt is not None:
        batch_size = len(prompt)
    else:
        batch_size = prompt_embeds.shape[0]

    if prompt_embeds is None:
        prompt_2 = prompt_2 or prompt
        prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2

        prompt_3 = prompt_3 or prompt
        prompt_3 = [prompt_3] if isinstance(prompt_3, str) else prompt_3

        prompt_embed, pooled_prompt_embed = self._get_clip_prompt_embeds(
            prompt=prompt,
            num_images_per_prompt=num_images_per_prompt,
            clip_skip=clip_skip,
            clip_model_index=0,
        )
        prompt_2_embed, pooled_prompt_2_embed = self._get_clip_prompt_embeds(
            prompt=prompt_2,
            num_images_per_prompt=num_images_per_prompt,
            clip_skip=clip_skip,
            clip_model_index=1,
        )
        clip_prompt_embeds = ops.cat([prompt_embed, prompt_2_embed], axis=-1)

        t5_prompt_embed = self._get_t5_prompt_embeds(
            prompt=prompt_3,
            num_images_per_prompt=num_images_per_prompt,
            max_sequence_length=max_sequence_length,
        )

        clip_prompt_embeds = pad(clip_prompt_embeds, (0, t5_prompt_embed.shape[-1] - clip_prompt_embeds.shape[-1]))

        prompt_embeds = ops.cat([clip_prompt_embeds, t5_prompt_embed], axis=-2)
        pooled_prompt_embeds = ops.cat([pooled_prompt_embed, pooled_prompt_2_embed], axis=-1)

    if do_classifier_free_guidance and negative_prompt_embeds is None:
        negative_prompt = negative_prompt or ""
        negative_prompt_2 = negative_prompt_2 or negative_prompt
        negative_prompt_3 = negative_prompt_3 or negative_prompt

        # normalize str to list
        negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
        negative_prompt_2 = (
            batch_size * [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2
        )
        negative_prompt_3 = (
            batch_size * [negative_prompt_3] if isinstance(negative_prompt_3, str) else negative_prompt_3
        )

        if prompt is not None and type(prompt) is not type(negative_prompt):
            raise TypeError(
                f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
                f" {type(prompt)}."
            )
        elif batch_size != len(negative_prompt):
            raise ValueError(
                f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
                f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
                " the batch size of `prompt`."
            )

        negative_prompt_embed, negative_pooled_prompt_embed = self._get_clip_prompt_embeds(
            negative_prompt,
            num_images_per_prompt=num_images_per_prompt,
            clip_skip=None,
            clip_model_index=0,
        )
        negative_prompt_2_embed, negative_pooled_prompt_2_embed = self._get_clip_prompt_embeds(
            negative_prompt_2,
            num_images_per_prompt=num_images_per_prompt,
            clip_skip=None,
            clip_model_index=1,
        )
        negative_clip_prompt_embeds = ops.cat([negative_prompt_embed, negative_prompt_2_embed], axis=-1)

        t5_negative_prompt_embed = self._get_t5_prompt_embeds(
            prompt=negative_prompt_3,
            num_images_per_prompt=num_images_per_prompt,
            max_sequence_length=max_sequence_length,
        )

        negative_clip_prompt_embeds = pad(
            negative_clip_prompt_embeds,
            (0, t5_negative_prompt_embed.shape[-1] - negative_clip_prompt_embeds.shape[-1]),
        )

        negative_prompt_embeds = ops.cat([negative_clip_prompt_embeds, t5_negative_prompt_embed], axis=-2)
        negative_pooled_prompt_embeds = ops.cat(
            [negative_pooled_prompt_embed, negative_pooled_prompt_2_embed], axis=-1
        )

    if self.text_encoder is not None:
        if isinstance(self, SD3LoraLoaderMixin):
            # Retrieve the original scale by scaling back the LoRA layers
            unscale_lora_layers(self.text_encoder, lora_scale)

    if self.text_encoder_2 is not None:
        if isinstance(self, SD3LoraLoaderMixin):
            # Retrieve the original scale by scaling back the LoRA layers
            unscale_lora_layers(self.text_encoder_2, lora_scale)

    return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds

mindone.diffusers.StableDiffusion3Pipeline.prepare_ip_adapter_image_embeds(ip_adapter_image=None, ip_adapter_image_embeds=None, num_images_per_prompt=1, do_classifier_free_guidance=True)

Prepares image embeddings for use in the IP-Adapter.

Either ip_adapter_image or ip_adapter_image_embeds must be passed.

PARAMETER	DESCRIPTION
ip_adapter_image	The input image to extract features from for IP-Adapter. TYPE: `PipelineImageInput`, *optional* DEFAULT: None
ip_adapter_image_embeds	Precomputed image embeddings. TYPE: `ms.Tensor`, *optional* DEFAULT: None
num_images_per_prompt	Number of images that should be generated per prompt. TYPE: `int`, defaults to 1 DEFAULT: 1
do_classifier_free_guidance	Whether to use classifier free guidance or not. TYPE: `bool`, defaults to True DEFAULT: True

Source code in mindone/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py

def prepare_ip_adapter_image_embeds(
    self,
    ip_adapter_image: Optional[PipelineImageInput] = None,
    ip_adapter_image_embeds: Optional[ms.Tensor] = None,
    num_images_per_prompt: int = 1,
    do_classifier_free_guidance: bool = True,
) -> ms.Tensor:
    """Prepares image embeddings for use in the IP-Adapter.

    Either `ip_adapter_image` or `ip_adapter_image_embeds` must be passed.

    Args:
        ip_adapter_image (`PipelineImageInput`, *optional*):
            The input image to extract features from for IP-Adapter.
        ip_adapter_image_embeds (`ms.Tensor`, *optional*):
            Precomputed image embeddings.
        num_images_per_prompt (`int`, defaults to 1):
            Number of images that should be generated per prompt.
        do_classifier_free_guidance (`bool`, defaults to True):
            Whether to use classifier free guidance or not.
    """
    if ip_adapter_image_embeds is not None:
        if do_classifier_free_guidance:
            single_negative_image_embeds, single_image_embeds = ip_adapter_image_embeds.chunk(2)
        else:
            single_image_embeds = ip_adapter_image_embeds
    elif ip_adapter_image is not None:
        single_image_embeds = self.encode_image(ip_adapter_image)
        if do_classifier_free_guidance:
            single_negative_image_embeds = ops.zeros_like(single_image_embeds)
    else:
        raise ValueError("Neither `ip_adapter_image_embeds` or `ip_adapter_image_embeds` were provided.")

    image_embeds = ops.cat([single_image_embeds] * num_images_per_prompt, axis=0)

    if do_classifier_free_guidance:
        negative_image_embeds = ops.cat([single_negative_image_embeds] * num_images_per_prompt, axis=0)
        image_embeds = ops.cat([negative_image_embeds, image_embeds], axis=0)

    return image_embeds