HiDreamImage

HiDream-I1 by HiDream.ai

Tip

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

Available models

The following models are available for the HiDreamImagePipeline pipeline:

Model name	Description
HiDream-ai/HiDream-I1-Full	-
HiDream-ai/HiDream-I1-Dev	-
HiDream-ai/HiDream-I1-Fast	-

mindone.diffusers.HiDreamImagePipeline

Bases: DiffusionPipeline, HiDreamImageLoraLoaderMixin

Source code in mindone/diffusers/pipelines/hidream_image/pipeline_hidream_image.py

class HiDreamImagePipeline(DiffusionPipeline, HiDreamImageLoraLoaderMixin):
    model_cpu_offload_seq = "text_encoder->text_encoder_2->text_encoder_3->text_encoder_4->transformer->vae"
    _callback_tensor_inputs = ["latents", "prompt_embeds_t5", "prompt_embeds_llama3", "pooled_prompt_embeds"]

    def __init__(
        self,
        scheduler: FlowMatchEulerDiscreteScheduler,
        vae: AutoencoderKL,
        text_encoder: CLIPTextModelWithProjection,
        tokenizer: CLIPTokenizer,
        text_encoder_2: CLIPTextModelWithProjection,
        tokenizer_2: CLIPTokenizer,
        text_encoder_3: T5EncoderModel,
        tokenizer_3: T5Tokenizer,
        text_encoder_4: LlamaForCausalLM,
        tokenizer_4: PreTrainedTokenizerFast,
        transformer: HiDreamImageTransformer2DModel,
    ):
        super().__init__()

        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            text_encoder_2=text_encoder_2,
            text_encoder_3=text_encoder_3,
            text_encoder_4=text_encoder_4,
            tokenizer=tokenizer,
            tokenizer_2=tokenizer_2,
            tokenizer_3=tokenizer_3,
            tokenizer_4=tokenizer_4,
            scheduler=scheduler,
            transformer=transformer,
        )
        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
        )
        # HiDreamImage latents are turned into 2x2 patches and packed. This means the latent width and height has to be divisible
        # by the patch size. So the vae scale factor is multiplied by the patch size to account for this
        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor * 2)
        self.default_sample_size = 128
        if getattr(self, "tokenizer_4", None) is not None:
            self.tokenizer_4.pad_token = self.tokenizer_4.eos_token

    def _get_t5_prompt_embeds(
        self,
        prompt: Union[str, List[str]] = None,
        max_sequence_length: int = 128,
        dtype: Optional[ms.Type] = None,
    ):
        dtype = dtype or self.text_encoder_3.dtype

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

        text_inputs = self.tokenizer_3(
            prompt,
            padding="max_length",
            max_length=min(max_sequence_length, self.tokenizer_3.model_max_length),
            truncation=True,
            add_special_tokens=True,
            return_tensors="np",
        )
        text_input_ids = text_inputs.input_ids
        attention_mask = text_inputs.attention_mask
        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[:, min(max_sequence_length, self.tokenizer_3.model_max_length) - 1 : -1]
            )
            logger.warning(
                "The following part of your input was truncated because `max_sequence_length` is set to "
                f" {min(max_sequence_length, self.tokenizer_3.model_max_length)} tokens: {removed_text}"
            )

        prompt_embeds = self.text_encoder_3(ms.tensor(text_input_ids), attention_mask=ms.tensor(attention_mask))[0]
        prompt_embeds = prompt_embeds.to(dtype=dtype)
        return prompt_embeds

    def _get_clip_prompt_embeds(
        self,
        tokenizer,
        text_encoder,
        prompt: Union[str, List[str]],
        max_sequence_length: int = 128,
        dtype: Optional[ms.Type] = None,
    ):
        dtype = dtype or text_encoder.dtype

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

        text_inputs = tokenizer(
            prompt,
            padding="max_length",
            max_length=min(max_sequence_length, 218),
            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[:, 218 - 1 : -1])
            logger.warning(
                "The following part of your input was truncated because CLIP can only handle sequences up to"
                f" {218} tokens: {removed_text}"
            )
        prompt_embeds = text_encoder(ms.tensor(text_input_ids), output_hidden_states=True)

        # Use pooled output of CLIPTextModel
        prompt_embeds = prompt_embeds[0]
        prompt_embeds = prompt_embeds.to(dtype=dtype)
        return prompt_embeds

    def _get_llama3_prompt_embeds(
        self,
        prompt: Union[str, List[str]] = None,
        max_sequence_length: int = 128,
        dtype: Optional[ms.Type] = None,
    ):
        dtype = dtype or self.text_encoder_4.dtype

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

        text_inputs = self.tokenizer_4(
            prompt,
            padding="max_length",
            max_length=min(max_sequence_length, self.tokenizer_4.model_max_length),
            truncation=True,
            add_special_tokens=True,
            return_tensors="np",
        )
        text_input_ids = text_inputs.input_ids
        attention_mask = text_inputs.attention_mask
        untruncated_ids = self.tokenizer_4(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_4.batch_decode(
                untruncated_ids[:, min(max_sequence_length, self.tokenizer_4.model_max_length) - 1 : -1]
            )
            logger.warning(
                "The following part of your input was truncated because `max_sequence_length` is set to "
                f" {min(max_sequence_length, self.tokenizer_4.model_max_length)} tokens: {removed_text}"
            )

        outputs = self.text_encoder_4(
            ms.tensor(text_input_ids),
            attention_mask=ms.tensor(attention_mask),
            output_hidden_states=True,
            output_attentions=True,
        )

        # prompt_embeds = outputs.hidden_states[1:]
        prompt_embeds = outputs[1][1:]
        prompt_embeds = mint.stack(prompt_embeds, dim=0)
        return prompt_embeds

    def encode_prompt(
        self,
        prompt: Optional[Union[str, List[str]]] = None,
        prompt_2: Optional[Union[str, List[str]]] = None,
        prompt_3: Optional[Union[str, List[str]]] = None,
        prompt_4: Optional[Union[str, List[str]]] = None,
        dtype: Optional[ms.Type] = None,
        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,
        negative_prompt_4: Optional[Union[str, List[str]]] = None,
        prompt_embeds_t5: Optional[List[ms.Tensor]] = None,
        prompt_embeds_llama3: Optional[List[ms.Tensor]] = None,
        negative_prompt_embeds_t5: Optional[List[ms.Tensor]] = None,
        negative_prompt_embeds_llama3: Optional[List[ms.Tensor]] = None,
        pooled_prompt_embeds: Optional[ms.Tensor] = None,
        negative_pooled_prompt_embeds: Optional[ms.Tensor] = None,
        max_sequence_length: int = 128,
        lora_scale: Optional[float] = None,
    ):
        prompt = [prompt] if isinstance(prompt, str) else prompt
        if prompt is not None:
            batch_size = len(prompt)
        else:
            batch_size = pooled_prompt_embeds.shape[0]

        if pooled_prompt_embeds is None:
            pooled_prompt_embeds_1 = self._get_clip_prompt_embeds(
                self.tokenizer, self.text_encoder, prompt, max_sequence_length, dtype
            )

        if do_classifier_free_guidance and negative_pooled_prompt_embeds is None:
            negative_prompt = negative_prompt or ""
            negative_prompt = [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt

            if len(negative_prompt) > 1 and len(negative_prompt) != batch_size:
                raise ValueError(f"negative_prompt must be of length 1 or {batch_size}")

            negative_pooled_prompt_embeds_1 = self._get_clip_prompt_embeds(
                self.tokenizer, self.text_encoder, negative_prompt, max_sequence_length, dtype
            )

            if negative_pooled_prompt_embeds_1.shape[0] == 1 and batch_size > 1:
                negative_pooled_prompt_embeds_1 = negative_pooled_prompt_embeds_1.repeat(batch_size, 1)

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

            if len(prompt_2) > 1 and len(prompt_2) != batch_size:
                raise ValueError(f"prompt_2 must be of length 1 or {batch_size}")

            pooled_prompt_embeds_2 = self._get_clip_prompt_embeds(
                self.tokenizer_2, self.text_encoder_2, prompt_2, max_sequence_length, dtype
            )

            if pooled_prompt_embeds_2.shape[0] == 1 and batch_size > 1:
                pooled_prompt_embeds_2 = pooled_prompt_embeds_2.repeat(batch_size, 1)

        if do_classifier_free_guidance and negative_pooled_prompt_embeds is None:
            negative_prompt_2 = negative_prompt_2 or negative_prompt
            negative_prompt_2 = [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2

            if len(negative_prompt_2) > 1 and len(negative_prompt_2) != batch_size:
                raise ValueError(f"negative_prompt_2 must be of length 1 or {batch_size}")

            negative_pooled_prompt_embeds_2 = self._get_clip_prompt_embeds(
                self.tokenizer_2, self.text_encoder_2, negative_prompt_2, max_sequence_length, dtype
            )

            if negative_pooled_prompt_embeds_2.shape[0] == 1 and batch_size > 1:
                negative_pooled_prompt_embeds_2 = negative_pooled_prompt_embeds_2.repeat(batch_size, 1)

        if pooled_prompt_embeds is None:
            pooled_prompt_embeds = mint.cat([pooled_prompt_embeds_1, pooled_prompt_embeds_2], dim=-1)

        if do_classifier_free_guidance and negative_pooled_prompt_embeds is None:
            negative_pooled_prompt_embeds = mint.cat(
                [negative_pooled_prompt_embeds_1, negative_pooled_prompt_embeds_2], dim=-1
            )

        if prompt_embeds_t5 is None:
            prompt_3 = prompt_3 or prompt
            prompt_3 = [prompt_3] if isinstance(prompt_3, str) else prompt_3

            if len(prompt_3) > 1 and len(prompt_3) != batch_size:
                raise ValueError(f"prompt_3 must be of length 1 or {batch_size}")

            prompt_embeds_t5 = self._get_t5_prompt_embeds(prompt_3, max_sequence_length, dtype)

            if prompt_embeds_t5.shape[0] == 1 and batch_size > 1:
                prompt_embeds_t5 = prompt_embeds_t5.repeat(batch_size, 1, 1)

        if do_classifier_free_guidance and negative_prompt_embeds_t5 is None:
            negative_prompt_3 = negative_prompt_3 or negative_prompt
            negative_prompt_3 = [negative_prompt_3] if isinstance(negative_prompt_3, str) else negative_prompt_3

            if len(negative_prompt_3) > 1 and len(negative_prompt_3) != batch_size:
                raise ValueError(f"negative_prompt_3 must be of length 1 or {batch_size}")

            negative_prompt_embeds_t5 = self._get_t5_prompt_embeds(negative_prompt_3, max_sequence_length, dtype)

            if negative_prompt_embeds_t5.shape[0] == 1 and batch_size > 1:
                negative_prompt_embeds_t5 = negative_prompt_embeds_t5.repeat(batch_size, 1, 1)

        if prompt_embeds_llama3 is None:
            prompt_4 = prompt_4 or prompt
            prompt_4 = [prompt_4] if isinstance(prompt_4, str) else prompt_4

            if len(prompt_4) > 1 and len(prompt_4) != batch_size:
                raise ValueError(f"prompt_4 must be of length 1 or {batch_size}")

            prompt_embeds_llama3 = self._get_llama3_prompt_embeds(prompt_4, max_sequence_length, dtype)

            if prompt_embeds_llama3.shape[0] == 1 and batch_size > 1:
                prompt_embeds_llama3 = prompt_embeds_llama3.repeat(1, batch_size, 1, 1)

        if do_classifier_free_guidance and negative_prompt_embeds_llama3 is None:
            negative_prompt_4 = negative_prompt_4 or negative_prompt
            negative_prompt_4 = [negative_prompt_4] if isinstance(negative_prompt_4, str) else negative_prompt_4

            if len(negative_prompt_4) > 1 and len(negative_prompt_4) != batch_size:
                raise ValueError(f"negative_prompt_4 must be of length 1 or {batch_size}")

            negative_prompt_embeds_llama3 = self._get_llama3_prompt_embeds(
                negative_prompt_4, max_sequence_length, dtype
            )

            if negative_prompt_embeds_llama3.shape[0] == 1 and batch_size > 1:
                negative_prompt_embeds_llama3 = negative_prompt_embeds_llama3.repeat(1, batch_size, 1, 1)

        # duplicate pooled_prompt_embeds for each generation per prompt
        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt)
        pooled_prompt_embeds = pooled_prompt_embeds.view(batch_size * num_images_per_prompt, -1)

        # duplicate t5_prompt_embeds for batch_size and num_images_per_prompt
        bs_embed, seq_len, _ = prompt_embeds_t5.shape
        if bs_embed == 1 and batch_size > 1:
            prompt_embeds_t5 = prompt_embeds_t5.repeat(batch_size, 1, 1)
        elif bs_embed > 1 and bs_embed != batch_size:
            raise ValueError(f"cannot duplicate prompt_embeds_t5 of batch size {bs_embed}")
        prompt_embeds_t5 = prompt_embeds_t5.repeat(1, num_images_per_prompt, 1)
        prompt_embeds_t5 = prompt_embeds_t5.view(batch_size * num_images_per_prompt, seq_len, -1)

        # duplicate llama3_prompt_embeds for batch_size and num_images_per_prompt
        _, bs_embed, seq_len, dim = prompt_embeds_llama3.shape
        if bs_embed == 1 and batch_size > 1:
            prompt_embeds_llama3 = prompt_embeds_llama3.repeat(1, batch_size, 1, 1)
        elif bs_embed > 1 and bs_embed != batch_size:
            raise ValueError(f"cannot duplicate prompt_embeds_llama3 of batch size {bs_embed}")
        prompt_embeds_llama3 = prompt_embeds_llama3.repeat(1, 1, num_images_per_prompt, 1)
        prompt_embeds_llama3 = prompt_embeds_llama3.view(-1, batch_size * num_images_per_prompt, seq_len, dim)

        if do_classifier_free_guidance:
            # duplicate negative_pooled_prompt_embeds for batch_size and num_images_per_prompt
            bs_embed, seq_len = negative_pooled_prompt_embeds.shape
            if bs_embed == 1 and batch_size > 1:
                negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(batch_size, 1)
            elif bs_embed > 1 and bs_embed != batch_size:
                raise ValueError(f"cannot duplicate negative_pooled_prompt_embeds of batch size {bs_embed}")
            negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt)
            negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.view(batch_size * num_images_per_prompt, -1)

            # duplicate negative_t5_prompt_embeds for batch_size and num_images_per_prompt
            bs_embed, seq_len, _ = negative_prompt_embeds_t5.shape
            if bs_embed == 1 and batch_size > 1:
                negative_prompt_embeds_t5 = negative_prompt_embeds_t5.repeat(batch_size, 1, 1)
            elif bs_embed > 1 and bs_embed != batch_size:
                raise ValueError(f"cannot duplicate negative_prompt_embeds_t5 of batch size {bs_embed}")
            negative_prompt_embeds_t5 = negative_prompt_embeds_t5.repeat(1, num_images_per_prompt, 1)
            negative_prompt_embeds_t5 = negative_prompt_embeds_t5.view(batch_size * num_images_per_prompt, seq_len, -1)

            # duplicate negative_prompt_embeds_llama3 for batch_size and num_images_per_prompt
            _, bs_embed, seq_len, dim = negative_prompt_embeds_llama3.shape
            if bs_embed == 1 and batch_size > 1:
                negative_prompt_embeds_llama3 = negative_prompt_embeds_llama3.repeat(1, batch_size, 1, 1)
            elif bs_embed > 1 and bs_embed != batch_size:
                raise ValueError(f"cannot duplicate negative_prompt_embeds_llama3 of batch size {bs_embed}")
            negative_prompt_embeds_llama3 = negative_prompt_embeds_llama3.repeat(1, 1, num_images_per_prompt, 1)
            negative_prompt_embeds_llama3 = negative_prompt_embeds_llama3.view(
                -1, batch_size * num_images_per_prompt, seq_len, dim
            )

        return (
            prompt_embeds_t5,
            negative_prompt_embeds_t5,
            prompt_embeds_llama3,
            negative_prompt_embeds_llama3,
            pooled_prompt_embeds,
            negative_pooled_prompt_embeds,
        )

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

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

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

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

    def check_inputs(
        self,
        prompt,
        prompt_2,
        prompt_3,
        prompt_4,
        negative_prompt=None,
        negative_prompt_2=None,
        negative_prompt_3=None,
        negative_prompt_4=None,
        prompt_embeds_t5=None,
        prompt_embeds_llama3=None,
        negative_prompt_embeds_t5=None,
        negative_prompt_embeds_llama3=None,
        pooled_prompt_embeds=None,
        negative_pooled_prompt_embeds=None,
        callback_on_step_end_tensor_inputs=None,
    ):
        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]}"
            )

        if prompt is not None and pooled_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt`: {prompt} and `pooled_prompt_embeds`: {pooled_prompt_embeds}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt_2 is not None and pooled_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt_2`: {prompt_2} and `pooled_prompt_embeds`: {pooled_prompt_embeds}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt_3 is not None and prompt_embeds_t5 is not None:
            raise ValueError(
                f"Cannot forward both `prompt_3`: {prompt_3} and `prompt_embeds_t5`: {prompt_embeds_t5}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt_4 is not None and prompt_embeds_llama3 is not None:
            raise ValueError(
                f"Cannot forward both `prompt_4`: {prompt_4} and `prompt_embeds_llama3`: {prompt_embeds_llama3}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt is None and pooled_prompt_embeds is None:
            raise ValueError(
                "Provide either `prompt` or `pooled_prompt_embeds`. Cannot leave both `prompt` and `pooled_prompt_embeds` undefined."
            )
        elif prompt is None and prompt_embeds_t5 is None:
            raise ValueError(
                "Provide either `prompt` or `prompt_embeds_t5`. Cannot leave both `prompt` and `prompt_embeds_t5` undefined."
            )
        elif prompt is None and prompt_embeds_llama3 is None:
            raise ValueError(
                "Provide either `prompt` or `prompt_embeds_llama3`. Cannot leave both `prompt` and `prompt_embeds_llama3` 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)}")
        elif prompt_4 is not None and (not isinstance(prompt_4, str) and not isinstance(prompt_4, list)):
            raise ValueError(f"`prompt_4` has to be of type `str` or `list` but is {type(prompt_4)}")

        if negative_prompt is not None and negative_pooled_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_pooled_prompt_embeds`:"
                f" {negative_pooled_prompt_embeds}. Please make sure to only forward one of the two."
            )
        elif negative_prompt_2 is not None and negative_pooled_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} and `negative_pooled_prompt_embeds`:"
                f" {negative_pooled_prompt_embeds}. Please make sure to only forward one of the two."
            )
        elif negative_prompt_3 is not None and negative_prompt_embeds_t5 is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt_3`: {negative_prompt_3} and `negative_prompt_embeds_t5`:"
                f" {negative_prompt_embeds_t5}. Please make sure to only forward one of the two."
            )
        elif negative_prompt_4 is not None and negative_prompt_embeds_llama3 is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt_4`: {negative_prompt_4} and `negative_prompt_embeds_llama3`:"
                f" {negative_prompt_embeds_llama3}. Please make sure to only forward one of the two."
            )

        if pooled_prompt_embeds is not None and negative_pooled_prompt_embeds is not None:
            if pooled_prompt_embeds.shape != negative_pooled_prompt_embeds.shape:
                raise ValueError(
                    "`pooled_prompt_embeds` and `negative_pooled_prompt_embeds` must have the same shape when passed directly, but"
                    f" got: `pooled_prompt_embeds` {pooled_prompt_embeds.shape} != `negative_pooled_prompt_embeds`"
                    f" {negative_pooled_prompt_embeds.shape}."
                )
        if prompt_embeds_t5 is not None and negative_prompt_embeds_t5 is not None:
            if prompt_embeds_t5.shape != negative_prompt_embeds_t5.shape:
                raise ValueError(
                    "`prompt_embeds_t5` and `negative_prompt_embeds_t5` must have the same shape when passed directly, but"
                    f" got: `prompt_embeds_t5` {prompt_embeds_t5.shape} != `negative_prompt_embeds_t5`"
                    f" {negative_prompt_embeds_t5.shape}."
                )
        if prompt_embeds_llama3 is not None and negative_prompt_embeds_llama3 is not None:
            if prompt_embeds_llama3.shape != negative_prompt_embeds_llama3.shape:
                raise ValueError(
                    "`prompt_embeds_llama3` and `negative_prompt_embeds_llama3` must have the same shape when passed directly, but"
                    f" got: `prompt_embeds_llama3` {prompt_embeds_llama3.shape} != `negative_prompt_embeds_llama3`"
                    f" {negative_prompt_embeds_llama3.shape}."
                )

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

        shape = (batch_size, num_channels_latents, height, width)

        if latents is None:
            latents = randn_tensor(shape, generator=generator, dtype=dtype)
        else:
            if latents.shape != shape:
                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {shape}")
            latents = latents
        return latents

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

    @property
    def do_classifier_free_guidance(self):
        return self._guidance_scale > 1

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

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

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

    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,
        prompt_4: Optional[Union[str, List[str]]] = None,
        height: Optional[int] = None,
        width: Optional[int] = None,
        num_inference_steps: int = 50,
        sigmas: Optional[List[float]] = None,
        guidance_scale: float = 5.0,
        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,
        negative_prompt_4: 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_t5: Optional[ms.Tensor] = None,
        prompt_embeds_llama3: Optional[ms.Tensor] = None,
        negative_prompt_embeds_t5: Optional[ms.Tensor] = None,
        negative_prompt_embeds_llama3: Optional[ms.Tensor] = None,
        pooled_prompt_embeds: Optional[ms.Tensor] = None,
        negative_pooled_prompt_embeds: Optional[ms.Tensor] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = False,
        attention_kwargs: Optional[Dict[str, Any]] = None,
        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
        max_sequence_length: int = 128,
        **kwargs,
    ):
        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.
            prompt_4 (`str` or `List[str]`, *optional*):
                The prompt or prompts to be sent to `tokenizer_4` and `text_encoder_4`. 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 3.5):
                Guidance scale as defined in [Classifier-Free Diffusion
                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
                of [Imagen Paper](https://huggingface.co/papers/2205.11487). 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 `true_cfg_scale` is
                not greater 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_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 in all the text-encoders.
            negative_prompt_4 (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation to be sent to `tokenizer_4` and
                `text_encoder_4`. If not defined, `negative_prompt` is used in all the text-encoders.
            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 [np.random.Generator(s)](https://numpy.org/doc/stable/reference/random/generator.html)
                to make generation deterministic.
            latents (`ms.Tensor`, *optional*):
                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor will 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.
            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.flux.FluxPipelineOutput`] instead of a plain tuple.
            attention_kwargs (`dict`, *optional*):
                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
                `self.processor` in
                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
            callback_on_step_end (`Callable`, *optional*):
                A function that calls at the end of each denoising steps during the inference. The function is called
                with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
                callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
                `callback_on_step_end_tensor_inputs`.
            callback_on_step_end_tensor_inputs (`List`, *optional*):
                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
                `._callback_tensor_inputs` attribute of your pipeline class.
            max_sequence_length (`int` defaults to 128): Maximum sequence length to use with the `prompt`.

        Examples:

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

        prompt_embeds = kwargs.get("prompt_embeds", None)
        negative_prompt_embeds = kwargs.get("negative_prompt_embeds", None)

        if prompt_embeds is not None:
            deprecation_message = "The `prompt_embeds` argument is deprecated. \
                Please use `prompt_embeds_t5` and `prompt_embeds_llama3` instead."
            deprecate("prompt_embeds", "0.35.0", deprecation_message)
            prompt_embeds_t5 = prompt_embeds[0]
            prompt_embeds_llama3 = prompt_embeds[1]

        if negative_prompt_embeds is not None:
            deprecation_message = "The `negative_prompt_embeds` argument is deprecated. \
                Please use `negative_prompt_embeds_t5` and `negative_prompt_embeds_llama3` instead."
            deprecate("negative_prompt_embeds", "0.35.0", deprecation_message)
            negative_prompt_embeds_t5 = negative_prompt_embeds[0]
            negative_prompt_embeds_llama3 = negative_prompt_embeds[1]

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

        division = self.vae_scale_factor * 2
        S_max = (self.default_sample_size * self.vae_scale_factor) ** 2
        scale = S_max / (width * height)
        scale = math.sqrt(scale)
        width, height = int(width * scale // division * division), int(height * scale // division * division)

        # 1. Check inputs. Raise error if not correct
        self.check_inputs(
            prompt,
            prompt_2,
            prompt_3,
            prompt_4,
            negative_prompt=negative_prompt,
            negative_prompt_2=negative_prompt_2,
            negative_prompt_3=negative_prompt_3,
            negative_prompt_4=negative_prompt_4,
            prompt_embeds_t5=prompt_embeds_t5,
            prompt_embeds_llama3=prompt_embeds_llama3,
            negative_prompt_embeds_t5=negative_prompt_embeds_t5,
            negative_prompt_embeds_llama3=negative_prompt_embeds_llama3,
            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,
        )

        self._guidance_scale = guidance_scale
        self._attention_kwargs = 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)
        elif pooled_prompt_embeds is not None:
            batch_size = pooled_prompt_embeds.shape[0]

        # 3. Encode prompt
        lora_scale = self.attention_kwargs.get("scale", None) if self.attention_kwargs is not None else None
        (
            prompt_embeds_t5,
            negative_prompt_embeds_t5,
            prompt_embeds_llama3,
            negative_prompt_embeds_llama3,
            pooled_prompt_embeds,
            negative_pooled_prompt_embeds,
        ) = self.encode_prompt(
            prompt=prompt,
            prompt_2=prompt_2,
            prompt_3=prompt_3,
            prompt_4=prompt_4,
            negative_prompt=negative_prompt,
            negative_prompt_2=negative_prompt_2,
            negative_prompt_3=negative_prompt_3,
            negative_prompt_4=negative_prompt_4,
            do_classifier_free_guidance=self.do_classifier_free_guidance,
            prompt_embeds_t5=prompt_embeds_t5,
            prompt_embeds_llama3=prompt_embeds_llama3,
            negative_prompt_embeds_t5=negative_prompt_embeds_t5,
            negative_prompt_embeds_llama3=negative_prompt_embeds_llama3,
            pooled_prompt_embeds=pooled_prompt_embeds,
            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
            num_images_per_prompt=num_images_per_prompt,
            max_sequence_length=max_sequence_length,
            lora_scale=lora_scale,
        )

        if self.do_classifier_free_guidance:
            prompt_embeds_t5 = mint.cat([negative_prompt_embeds_t5, prompt_embeds_t5], dim=0)
            prompt_embeds_llama3 = mint.cat([negative_prompt_embeds_llama3, prompt_embeds_llama3], dim=1)
            pooled_prompt_embeds = mint.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds], dim=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,
            pooled_prompt_embeds.dtype,
            generator,
            latents,
        )

        # 5. Prepare timesteps
        mu = calculate_shift(self.transformer.max_seq)
        scheduler_kwargs = {"mu": mu}
        if isinstance(self.scheduler, UniPCMultistepScheduler):
            self.scheduler.set_timesteps(num_inference_steps)  # , shift=math.exp(mu))
            timesteps = self.scheduler.timesteps
        else:
            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. 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 = mint.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],))  # .to(latents.dtype)

                noise_pred = self.transformer(
                    hidden_states=latent_model_input,
                    timesteps=timestep,
                    encoder_hidden_states_t5=prompt_embeds_t5,
                    encoder_hidden_states_llama3=prompt_embeds_llama3,
                    pooled_embeds=pooled_prompt_embeds,
                    return_dict=False,
                )[0]
                noise_pred = -noise_pred

                # 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)

                # 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_t5 = callback_outputs.pop("prompt_embeds_t5", prompt_embeds_t5)
                    prompt_embeds_llama3 = callback_outputs.pop("prompt_embeds_llama3", prompt_embeds_llama3)
                    pooled_prompt_embeds = callback_outputs.pop("pooled_prompt_embeds", 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 output_type == "latent":
            image = latents

        else:
            latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor

            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 HiDreamImagePipelineOutput(images=image)

mindone.diffusers.HiDreamImagePipeline.__call__(prompt=None, prompt_2=None, prompt_3=None, prompt_4=None, height=None, width=None, num_inference_steps=50, sigmas=None, guidance_scale=5.0, negative_prompt=None, negative_prompt_2=None, negative_prompt_3=None, negative_prompt_4=None, num_images_per_prompt=1, generator=None, latents=None, prompt_embeds_t5=None, prompt_embeds_llama3=None, negative_prompt_embeds_t5=None, negative_prompt_embeds_llama3=None, pooled_prompt_embeds=None, negative_pooled_prompt_embeds=None, output_type='pil', return_dict=False, attention_kwargs=None, callback_on_step_end=None, callback_on_step_end_tensor_inputs=['latents'], max_sequence_length=128, **kwargs)

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
prompt_4	The prompt or prompts to be sent to tokenizer_4 and text_encoder_4. 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: 50
sigmas	Custom sigmas to use for the denoising process with schedulers which support a sigmas argument in their set_timesteps method. If not defined, the default behavior when num_inference_steps is passed will be used. TYPE: `List[float]`, *optional* DEFAULT: None
guidance_scale	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 3.5 DEFAULT: 5.0
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 true_cfg_scale is not greater 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_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 in all the text-encoders. TYPE: `str` or `List[str]`, *optional* DEFAULT: None
negative_prompt_4	The prompt or prompts not to guide the image generation to be sent to tokenizer_4 and text_encoder_4. If not defined, negative_prompt is used in all the text-encoders. 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 np.random.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*
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*
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
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.flux.FluxPipelineOutput] instead of a plain tuple. TYPE: `bool`, *optional*, defaults to `False` DEFAULT: False
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 128 DEFAULT: 128

RETURNS	DESCRIPTION
	[~pipelines.hidream_image.HiDreamImagePipelineOutput] or tuple:
	[~pipelines.hidream_image.HiDreamImagePipelineOutput] 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/hidream_image/pipeline_hidream_image.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,
    prompt_4: Optional[Union[str, List[str]]] = None,
    height: Optional[int] = None,
    width: Optional[int] = None,
    num_inference_steps: int = 50,
    sigmas: Optional[List[float]] = None,
    guidance_scale: float = 5.0,
    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,
    negative_prompt_4: 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_t5: Optional[ms.Tensor] = None,
    prompt_embeds_llama3: Optional[ms.Tensor] = None,
    negative_prompt_embeds_t5: Optional[ms.Tensor] = None,
    negative_prompt_embeds_llama3: Optional[ms.Tensor] = None,
    pooled_prompt_embeds: Optional[ms.Tensor] = None,
    negative_pooled_prompt_embeds: Optional[ms.Tensor] = None,
    output_type: Optional[str] = "pil",
    return_dict: bool = False,
    attention_kwargs: Optional[Dict[str, Any]] = None,
    callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
    callback_on_step_end_tensor_inputs: List[str] = ["latents"],
    max_sequence_length: int = 128,
    **kwargs,
):
    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.
        prompt_4 (`str` or `List[str]`, *optional*):
            The prompt or prompts to be sent to `tokenizer_4` and `text_encoder_4`. 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 3.5):
            Guidance scale as defined in [Classifier-Free Diffusion
            Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
            of [Imagen Paper](https://huggingface.co/papers/2205.11487). 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 `true_cfg_scale` is
            not greater 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_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 in all the text-encoders.
        negative_prompt_4 (`str` or `List[str]`, *optional*):
            The prompt or prompts not to guide the image generation to be sent to `tokenizer_4` and
            `text_encoder_4`. If not defined, `negative_prompt` is used in all the text-encoders.
        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 [np.random.Generator(s)](https://numpy.org/doc/stable/reference/random/generator.html)
            to make generation deterministic.
        latents (`ms.Tensor`, *optional*):
            Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
            generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
            tensor will 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.
        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.flux.FluxPipelineOutput`] instead of a plain tuple.
        attention_kwargs (`dict`, *optional*):
            A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
            `self.processor` in
            [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
        callback_on_step_end (`Callable`, *optional*):
            A function that calls at the end of each denoising steps during the inference. The function is called
            with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
            callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
            `callback_on_step_end_tensor_inputs`.
        callback_on_step_end_tensor_inputs (`List`, *optional*):
            The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
            will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
            `._callback_tensor_inputs` attribute of your pipeline class.
        max_sequence_length (`int` defaults to 128): Maximum sequence length to use with the `prompt`.

    Examples:

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

    prompt_embeds = kwargs.get("prompt_embeds", None)
    negative_prompt_embeds = kwargs.get("negative_prompt_embeds", None)

    if prompt_embeds is not None:
        deprecation_message = "The `prompt_embeds` argument is deprecated. \
            Please use `prompt_embeds_t5` and `prompt_embeds_llama3` instead."
        deprecate("prompt_embeds", "0.35.0", deprecation_message)
        prompt_embeds_t5 = prompt_embeds[0]
        prompt_embeds_llama3 = prompt_embeds[1]

    if negative_prompt_embeds is not None:
        deprecation_message = "The `negative_prompt_embeds` argument is deprecated. \
            Please use `negative_prompt_embeds_t5` and `negative_prompt_embeds_llama3` instead."
        deprecate("negative_prompt_embeds", "0.35.0", deprecation_message)
        negative_prompt_embeds_t5 = negative_prompt_embeds[0]
        negative_prompt_embeds_llama3 = negative_prompt_embeds[1]

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

    division = self.vae_scale_factor * 2
    S_max = (self.default_sample_size * self.vae_scale_factor) ** 2
    scale = S_max / (width * height)
    scale = math.sqrt(scale)
    width, height = int(width * scale // division * division), int(height * scale // division * division)

    # 1. Check inputs. Raise error if not correct
    self.check_inputs(
        prompt,
        prompt_2,
        prompt_3,
        prompt_4,
        negative_prompt=negative_prompt,
        negative_prompt_2=negative_prompt_2,
        negative_prompt_3=negative_prompt_3,
        negative_prompt_4=negative_prompt_4,
        prompt_embeds_t5=prompt_embeds_t5,
        prompt_embeds_llama3=prompt_embeds_llama3,
        negative_prompt_embeds_t5=negative_prompt_embeds_t5,
        negative_prompt_embeds_llama3=negative_prompt_embeds_llama3,
        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,
    )

    self._guidance_scale = guidance_scale
    self._attention_kwargs = 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)
    elif pooled_prompt_embeds is not None:
        batch_size = pooled_prompt_embeds.shape[0]

    # 3. Encode prompt
    lora_scale = self.attention_kwargs.get("scale", None) if self.attention_kwargs is not None else None
    (
        prompt_embeds_t5,
        negative_prompt_embeds_t5,
        prompt_embeds_llama3,
        negative_prompt_embeds_llama3,
        pooled_prompt_embeds,
        negative_pooled_prompt_embeds,
    ) = self.encode_prompt(
        prompt=prompt,
        prompt_2=prompt_2,
        prompt_3=prompt_3,
        prompt_4=prompt_4,
        negative_prompt=negative_prompt,
        negative_prompt_2=negative_prompt_2,
        negative_prompt_3=negative_prompt_3,
        negative_prompt_4=negative_prompt_4,
        do_classifier_free_guidance=self.do_classifier_free_guidance,
        prompt_embeds_t5=prompt_embeds_t5,
        prompt_embeds_llama3=prompt_embeds_llama3,
        negative_prompt_embeds_t5=negative_prompt_embeds_t5,
        negative_prompt_embeds_llama3=negative_prompt_embeds_llama3,
        pooled_prompt_embeds=pooled_prompt_embeds,
        negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
        num_images_per_prompt=num_images_per_prompt,
        max_sequence_length=max_sequence_length,
        lora_scale=lora_scale,
    )

    if self.do_classifier_free_guidance:
        prompt_embeds_t5 = mint.cat([negative_prompt_embeds_t5, prompt_embeds_t5], dim=0)
        prompt_embeds_llama3 = mint.cat([negative_prompt_embeds_llama3, prompt_embeds_llama3], dim=1)
        pooled_prompt_embeds = mint.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds], dim=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,
        pooled_prompt_embeds.dtype,
        generator,
        latents,
    )

    # 5. Prepare timesteps
    mu = calculate_shift(self.transformer.max_seq)
    scheduler_kwargs = {"mu": mu}
    if isinstance(self.scheduler, UniPCMultistepScheduler):
        self.scheduler.set_timesteps(num_inference_steps)  # , shift=math.exp(mu))
        timesteps = self.scheduler.timesteps
    else:
        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. 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 = mint.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],))  # .to(latents.dtype)

            noise_pred = self.transformer(
                hidden_states=latent_model_input,
                timesteps=timestep,
                encoder_hidden_states_t5=prompt_embeds_t5,
                encoder_hidden_states_llama3=prompt_embeds_llama3,
                pooled_embeds=pooled_prompt_embeds,
                return_dict=False,
            )[0]
            noise_pred = -noise_pred

            # 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)

            # 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_t5 = callback_outputs.pop("prompt_embeds_t5", prompt_embeds_t5)
                prompt_embeds_llama3 = callback_outputs.pop("prompt_embeds_llama3", prompt_embeds_llama3)
                pooled_prompt_embeds = callback_outputs.pop("pooled_prompt_embeds", 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 output_type == "latent":
        image = latents

    else:
        latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor

        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 HiDreamImagePipelineOutput(images=image)

mindone.diffusers.HiDreamImagePipeline.disable_vae_slicing()

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

Source code in mindone/diffusers/pipelines/hidream_image/pipeline_hidream_image.py

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

mindone.diffusers.HiDreamImagePipeline.disable_vae_tiling()

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

Source code in mindone/diffusers/pipelines/hidream_image/pipeline_hidream_image.py

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

mindone.diffusers.HiDreamImagePipeline.enable_vae_slicing()

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

Source code in mindone/diffusers/pipelines/hidream_image/pipeline_hidream_image.py

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

mindone.diffusers.HiDreamImagePipeline.enable_vae_tiling()

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

Source code in mindone/diffusers/pipelines/hidream_image/pipeline_hidream_image.py

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

mindone.diffusers.pipelines.hidream_image.pipeline_output.HiDreamImagePipelineOutput dataclass

Bases: BaseOutput

Output class for HiDreamImage pipelines.

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

@dataclass
class HiDreamImagePipelineOutput(BaseOutput):
    """
    Output class for HiDreamImage pipelines.

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

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