Refactor environment variable assignments in workflow

* Detect 2.0 vs 2.1 ZImageControlNetModel

* Possibility of control_noise_refiner being removed

.github/workflows/mirror_community_pipeline.yml

@@ -24,7 +24,6 @@ jobs:
   mirror_community_pipeline:
     env:
       SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK_URL_COMMUNITY_MIRROR }}
-
     runs-on: ubuntu-22.04
     steps:
       # Checkout to correct ref
@@ -39,25 +38,28 @@ jobs:
       #     If ref is 'refs/heads/main' => set 'main'
       #     Else it must be a tag => set {tag}
       - name: Set checkout_ref and path_in_repo
+          EVENT_NAME: ${{ github.event_name }}
+          EVENT_INPUT_REF: ${{ github.event.inputs.ref }}
+          GITHUB_REF: ${{ github.ref }}
         run: |
-          if [ "${{ github.event_name }}" == "workflow_dispatch" ]; then
-            if [ -z "${{ github.event.inputs.ref }}" ]; then
+          if [ "$EVENT_NAME" == "workflow_dispatch" ]; then
+            if [ -z "$EVENT_INPUT_REF" ]; then
               echo "Error: Missing ref input"
               exit 1
-            elif [ "${{ github.event.inputs.ref }}" == "main" ]; then
+            elif [ "$EVENT_INPUT_REF" == "main" ]; then
               echo "CHECKOUT_REF=refs/heads/main" >> $GITHUB_ENV
               echo "PATH_IN_REPO=main" >> $GITHUB_ENV
             else
-              echo "CHECKOUT_REF=refs/tags/${{ github.event.inputs.ref }}" >> $GITHUB_ENV
-              echo "PATH_IN_REPO=${{ github.event.inputs.ref }}" >> $GITHUB_ENV
+              echo "CHECKOUT_REF=refs/tags/$EVENT_INPUT_REF" >> $GITHUB_ENV
+              echo "PATH_IN_REPO=$EVENT_INPUT_REF" >> $GITHUB_ENV
             fi
-          elif [ "${{ github.ref }}" == "refs/heads/main" ]; then
-            echo "CHECKOUT_REF=${{ github.ref }}" >> $GITHUB_ENV
+          elif [ "$GITHUB_REF" == "refs/heads/main" ]; then
+            echo "CHECKOUT_REF=$GITHUB_REF" >> $GITHUB_ENV
             echo "PATH_IN_REPO=main" >> $GITHUB_ENV
           else
             # e.g. refs/tags/v0.28.1 -> v0.28.1
-            echo "CHECKOUT_REF=${{ github.ref }}" >> $GITHUB_ENV
-            echo "PATH_IN_REPO=$(echo ${{ github.ref }} | sed 's/^refs\/tags\///')" >> $GITHUB_ENV
+            echo "CHECKOUT_REF=$GITHUB_REF" >> $GITHUB_ENV
+            echo "PATH_IN_REPO=$(echo $GITHUB_REF | sed 's/^refs\/tags\///')" >> $GITHUB_ENV
           fi
       - name: Print env vars
         run: |
@@ -99,4 +101,4 @@ jobs:
       - name: Report failure status
         if: ${{ failure() }}
         run: |
-          pip install requests && python utils/notify_community_pipelines_mirror.py --status=failure
+          pip install requests && python utils/notify_community_pipelines_mirror.py --status=failure

docs/source/en/training/distributed_inference.md

@@ -263,8 +263,8 @@ def main():
     world_size = dist.get_world_size()
 
     pipeline = DiffusionPipeline.from_pretrained(
-        "black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16, device_map=device
-    )
+        "black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16
+    ).to(device)
     pipeline.transformer.set_attention_backend("_native_cudnn")
 
     cp_config = ContextParallelConfig(ring_degree=world_size)

examples/community/pipeline_z_image_differential_img2img.py

@@ -0,0 +1,844 @@
+# Copyright 2025 Alibaba Z-Image Team and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Union
+
+import torch
+from transformers import AutoTokenizer, PreTrainedModel
+
+from diffusers.image_processor import PipelineImageInput, VaeImageProcessor
+from diffusers.loaders import FromSingleFileMixin, ZImageLoraLoaderMixin
+from diffusers.models.autoencoders import AutoencoderKL
+from diffusers.models.transformers import ZImageTransformer2DModel
+from diffusers.pipelines.pipeline_utils import DiffusionPipeline
+from diffusers.pipelines.z_image.pipeline_output import ZImagePipelineOutput
+from diffusers.schedulers import FlowMatchEulerDiscreteScheduler
+from diffusers.utils import logging, replace_example_docstring
+from diffusers.utils.torch_utils import randn_tensor
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```py
+        >>> import torch
+        >>> from pipeline_z_image_differential_img2img import ZImageDifferentialImg2ImgPipeline
+        >>> from diffusers.utils import load_image
+
+        >>> pipe = ZImageDifferentialImg2ImgPipeline.from_pretrained("Z-a-o/Z-Image-Turbo", torch_dtype=torch.bfloat16)
+        >>> pipe.to("cuda")
+
+        >>> init_image = load_image(
+        >>>     "https://github.com/exx8/differential-diffusion/blob/main/assets/input.jpg?raw=true",
+        >>> )
+
+        >>> mask = load_image(
+        >>>     "https://github.com/exx8/differential-diffusion/blob/main/assets/map.jpg?raw=true",
+        >>> )
+
+        >>> prompt = "painting of a mountain landscape with a meadow and a forest, meadow background, anime countryside landscape, anime nature wallpap, anime landscape wallpaper, studio ghibli landscape, anime landscape, mountain behind meadow, anime background art, studio ghibli environment, background of flowery hill, anime beautiful peace scene, forrest background, anime scenery, landscape background, background art, anime scenery concept art"
+
+        >>> image = pipe(
+        ...     prompt,
+        ...     image=init_image,
+        ...     mask_image=mask,
+        ...     strength=0.75,
+        ...     num_inference_steps=9,
+        ...     guidance_scale=0.0,
+        ...     generator=torch.Generator("cuda").manual_seed(41),
+        ... ).images[0]
+        >>> image.save("image.png")
+        ```
+"""
+
+
+# Copied from diffusers.pipelines.flux.pipeline_flux.calculate_shift
+def calculate_shift(
+    image_seq_len,
+    base_seq_len: int = 256,
+    max_seq_len: int = 4096,
+    base_shift: float = 0.5,
+    max_shift: float = 1.15,
+):
+    m = (max_shift - base_shift) / (max_seq_len - base_seq_len)
+    b = base_shift - m * base_seq_len
+    mu = image_seq_len * m + b
+    return mu
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+    encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+    if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+        return encoder_output.latent_dist.sample(generator)
+    elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+        return encoder_output.latent_dist.mode()
+    elif hasattr(encoder_output, "latents"):
+        return encoder_output.latents
+    else:
+        raise AttributeError("Could not access latents of provided encoder_output")
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
+    **kwargs,
+):
+    r"""
+    Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+    custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+    Args:
+        scheduler (`SchedulerMixin`):
+            The scheduler to get timesteps from.
+        num_inference_steps (`int`):
+            The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+            must be `None`.
+        device (`str` or `torch.device`, *optional*):
+            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        timesteps (`List[int]`, *optional*):
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
+
+    Returns:
+        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+        second element is the number of inference steps.
+    """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+    if timesteps is not None:
+        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accepts_timesteps:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" timestep schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
+class ZImageDifferentialImg2ImgPipeline(DiffusionPipeline, ZImageLoraLoaderMixin, FromSingleFileMixin):
+    r"""
+    The ZImage pipeline for image-to-image generation.
+
+    Args:
+        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 ([`PreTrainedModel`]):
+            A text encoder model to encode text prompts.
+        tokenizer ([`AutoTokenizer`]):
+            A tokenizer to tokenize text prompts.
+        transformer ([`ZImageTransformer2DModel`]):
+            A ZImage transformer model to denoise the encoded image latents.
+    """
+
+    model_cpu_offload_seq = "text_encoder->transformer->vae"
+    _optional_components = []
+    _callback_tensor_inputs = ["latents", "prompt_embeds"]
+
+    def __init__(
+        self,
+        scheduler: FlowMatchEulerDiscreteScheduler,
+        vae: AutoencoderKL,
+        text_encoder: PreTrainedModel,
+        tokenizer: AutoTokenizer,
+        transformer: ZImageTransformer2DModel,
+    ):
+        super().__init__()
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            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
+        )
+        latent_channels = self.vae.config.latent_channels if getattr(self, "vae", None) else 16
+        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor * 2)
+
+        self.mask_processor = VaeImageProcessor(
+            vae_scale_factor=self.vae_scale_factor,
+            vae_latent_channels=latent_channels,
+            do_normalize=False,
+            do_binarize=False,
+            do_convert_grayscale=True,
+        )
+
+    # Copied from diffusers.pipelines.z_image.pipeline_z_image.ZImagePipeline.encode_prompt
+    def encode_prompt(
+        self,
+        prompt: Union[str, List[str]],
+        device: Optional[torch.device] = None,
+        do_classifier_free_guidance: bool = True,
+        negative_prompt: Optional[Union[str, List[str]]] = None,
+        prompt_embeds: Optional[List[torch.FloatTensor]] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        max_sequence_length: int = 512,
+    ):
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        prompt_embeds = self._encode_prompt(
+            prompt=prompt,
+            device=device,
+            prompt_embeds=prompt_embeds,
+            max_sequence_length=max_sequence_length,
+        )
+
+        if do_classifier_free_guidance:
+            if negative_prompt is None:
+                negative_prompt = ["" for _ in prompt]
+            else:
+                negative_prompt = [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+            assert len(prompt) == len(negative_prompt)
+            negative_prompt_embeds = self._encode_prompt(
+                prompt=negative_prompt,
+                device=device,
+                prompt_embeds=negative_prompt_embeds,
+                max_sequence_length=max_sequence_length,
+            )
+        else:
+            negative_prompt_embeds = []
+        return prompt_embeds, negative_prompt_embeds
+
+    # Copied from diffusers.pipelines.z_image.pipeline_z_image.ZImagePipeline._encode_prompt
+    def _encode_prompt(
+        self,
+        prompt: Union[str, List[str]],
+        device: Optional[torch.device] = None,
+        prompt_embeds: Optional[List[torch.FloatTensor]] = None,
+        max_sequence_length: int = 512,
+    ) -> List[torch.FloatTensor]:
+        device = device or self._execution_device
+
+        if prompt_embeds is not None:
+            return prompt_embeds
+
+        if isinstance(prompt, str):
+            prompt = [prompt]
+
+        for i, prompt_item in enumerate(prompt):
+            messages = [
+                {"role": "user", "content": prompt_item},
+            ]
+            prompt_item = self.tokenizer.apply_chat_template(
+                messages,
+                tokenize=False,
+                add_generation_prompt=True,
+                enable_thinking=True,
+            )
+            prompt[i] = prompt_item
+
+        text_inputs = self.tokenizer(
+            prompt,
+            padding="max_length",
+            max_length=max_sequence_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+
+        text_input_ids = text_inputs.input_ids.to(device)
+        prompt_masks = text_inputs.attention_mask.to(device).bool()
+
+        prompt_embeds = self.text_encoder(
+            input_ids=text_input_ids,
+            attention_mask=prompt_masks,
+            output_hidden_states=True,
+        ).hidden_states[-2]
+
+        embeddings_list = []
+
+        for i in range(len(prompt_embeds)):
+            embeddings_list.append(prompt_embeds[i][prompt_masks[i]])
+
+        return embeddings_list
+
+    # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img.StableDiffusion3Img2ImgPipeline.get_timesteps
+    def get_timesteps(self, num_inference_steps, strength, device):
+        # get the original timestep using init_timestep
+        init_timestep = min(num_inference_steps * strength, num_inference_steps)
+
+        t_start = int(max(num_inference_steps - init_timestep, 0))
+        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+        if hasattr(self.scheduler, "set_begin_index"):
+            self.scheduler.set_begin_index(t_start * self.scheduler.order)
+
+        return timesteps, num_inference_steps - t_start
+
+    @staticmethod
+    def _prepare_latent_image_ids(batch_size, height, width, device, dtype):
+        latent_image_ids = torch.zeros(height // 2, width // 2, 3)
+        latent_image_ids[..., 1] = latent_image_ids[..., 1] + torch.arange(height // 2)[:, None]
+        latent_image_ids[..., 2] = latent_image_ids[..., 2] + torch.arange(width // 2)[None, :]
+
+        latent_image_id_height, latent_image_id_width, latent_image_id_channels = latent_image_ids.shape
+
+        latent_image_ids = latent_image_ids.reshape(
+            latent_image_id_height * latent_image_id_width, latent_image_id_channels
+        )
+
+        return latent_image_ids.to(device=device, dtype=dtype)
+
+    def prepare_latents(
+        self,
+        image,
+        timestep,
+        batch_size,
+        num_channels_latents,
+        height,
+        width,
+        dtype,
+        device,
+        generator,
+        latents=None,
+    ):
+        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)
+        latent_image_ids = self._prepare_latent_image_ids(batch_size, height, width, device, dtype)
+
+        if latents is not None:
+            return latents.to(device=device, dtype=dtype)
+
+        # Encode the input image
+        image = image.to(device=device, dtype=dtype)
+        if image.shape[1] != num_channels_latents:
+            if isinstance(generator, list):
+                image_latents = [
+                    retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i])
+                    for i in range(image.shape[0])
+                ]
+                image_latents = torch.cat(image_latents, dim=0)
+            else:
+                image_latents = retrieve_latents(self.vae.encode(image), generator=generator)
+
+            # Apply scaling (inverse of decoding: decode does latents/scaling_factor + shift_factor)
+            image_latents = (image_latents - self.vae.config.shift_factor) * self.vae.config.scaling_factor
+        else:
+            image_latents = image
+
+        # Handle batch size expansion
+        if batch_size > image_latents.shape[0] and batch_size % image_latents.shape[0] == 0:
+            additional_image_per_prompt = batch_size // image_latents.shape[0]
+            image_latents = torch.cat([image_latents] * additional_image_per_prompt, dim=0)
+        elif batch_size > image_latents.shape[0] and batch_size % image_latents.shape[0] != 0:
+            raise ValueError(
+                f"Cannot duplicate `image` of batch size {image_latents.shape[0]} to {batch_size} text prompts."
+            )
+
+        # Add noise using flow matching scale_noise
+        noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+        latents = self.scheduler.scale_noise(image_latents, timestep, noise)
+
+        return latents, noise, image_latents, latent_image_ids
+
+    def prepare_mask_latents(
+        self,
+        mask,
+        masked_image,
+        batch_size,
+        num_images_per_prompt,
+        height,
+        width,
+        dtype,
+        device,
+        generator,
+    ):
+        height = 2 * (int(height) // (self.vae_scale_factor * 2))
+        width = 2 * (int(width) // (self.vae_scale_factor * 2))
+        # resize the mask to latents shape as we concatenate the mask to the latents
+        # we do that before converting to dtype to avoid breaking in case we're using cpu_offload
+        # and half precision
+        mask = torch.nn.functional.interpolate(mask, size=(height, width))
+        mask = mask.to(device=device, dtype=dtype)
+
+        batch_size = batch_size * num_images_per_prompt
+
+        masked_image = masked_image.to(device=device, dtype=dtype)
+
+        if masked_image.shape[1] == 16:
+            masked_image_latents = masked_image
+        else:
+            masked_image_latents = retrieve_latents(self.vae.encode(masked_image), generator=generator)
+
+        masked_image_latents = (masked_image_latents - self.vae.config.shift_factor) * self.vae.config.scaling_factor
+
+        # duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method
+        if mask.shape[0] < batch_size:
+            if not batch_size % mask.shape[0] == 0:
+                raise ValueError(
+                    "The passed mask and the required batch size don't match. Masks are supposed to be duplicated to"
+                    f" a total batch size of {batch_size}, but {mask.shape[0]} masks were passed. Make sure the number"
+                    " of masks that you pass is divisible by the total requested batch size."
+                )
+            mask = mask.repeat(batch_size // mask.shape[0], 1, 1, 1)
+        if masked_image_latents.shape[0] < batch_size:
+            if not batch_size % masked_image_latents.shape[0] == 0:
+                raise ValueError(
+                    "The passed images and the required batch size don't match. Images are supposed to be duplicated"
+                    f" to a total batch size of {batch_size}, but {masked_image_latents.shape[0]} images were passed."
+                    " Make sure the number of images that you pass is divisible by the total requested batch size."
+                )
+            masked_image_latents = masked_image_latents.repeat(batch_size // masked_image_latents.shape[0], 1, 1, 1)
+
+        # aligning device to prevent device errors when concating it with the latent model input
+        masked_image_latents = masked_image_latents.to(device=device, dtype=dtype)
+
+        return mask, masked_image_latents
+
+    @property
+    def guidance_scale(self):
+        return self._guidance_scale
+
+    @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
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        image: PipelineImageInput = None,
+        mask_image: PipelineImageInput = None,
+        strength: float = 0.6,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_inference_steps: int = 50,
+        sigmas: Optional[List[float]] = None,
+        guidance_scale: float = 5.0,
+        cfg_normalization: bool = False,
+        cfg_truncation: float = 1.0,
+        negative_prompt: Optional[Union[str, List[str]]] = None,
+        num_images_per_prompt: Optional[int] = 1,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[List[torch.FloatTensor]] = None,
+        negative_prompt_embeds: Optional[List[torch.FloatTensor]] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
+        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        max_sequence_length: int = 512,
+    ):
+        r"""
+        Function invoked when calling the pipeline for image-to-image 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.
+            image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`):
+                `Image`, numpy array or tensor representing an image batch to be used as the starting point. For both
+                numpy array and pytorch tensor, the expected value range is between `[0, 1]`. If it's a tensor or a
+                list of tensors, the expected shape should be `(B, C, H, W)` or `(C, H, W)`. If it is a numpy array or
+                a list of arrays, the expected shape should be `(B, H, W, C)` or `(H, W, C)`.
+            mask_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`):
+                `Image`, numpy array or tensor representing an image batch to mask `image`. Black pixels in the mask
+                are repainted while white pixels are preserved. If `mask_image` is a PIL image, it is converted to a
+                single channel (luminance) before use. If it's a numpy array or pytorch tensor, it should contain one
+                color channel (L) instead of 3, so the expected shape for pytorch tensor would be `(B, 1, H, W)`, `(B,
+                H, W)`, `(1, H, W)`, `(H, W)`. And for numpy array would be for `(B, H, W, 1)`, `(B, H, W)`, `(H, W,
+                1)`, or `(H, W)`.
+            strength (`float`, *optional*, defaults to 0.6):
+                Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a
+                starting point and more noise is added the higher the `strength`. The number of denoising steps depends
+                on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising
+                process runs for the full number of iterations specified in `num_inference_steps`. A value of 1
+                essentially ignores `image`.
+            height (`int`, *optional*, defaults to 1024):
+                The height in pixels of the generated image. If not provided, uses the input image height.
+            width (`int`, *optional*, defaults to 1024):
+                The width in pixels of the generated image. If not provided, uses the input image width.
+            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 5.0):
+                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.
+            cfg_normalization (`bool`, *optional*, defaults to False):
+                Whether to apply configuration normalization.
+            cfg_truncation (`float`, *optional*, defaults to 1.0):
+                The truncation value for configuration.
+            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`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+                to make generation deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will be generated by sampling using the supplied random `generator`.
+            prompt_embeds (`List[torch.FloatTensor]`, *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 (`List[torch.FloatTensor]`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.ZImagePipelineOutput`] 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`, *optional*, defaults to 512):
+                Maximum sequence length to use with the `prompt`.
+
+        Examples:
+
+        Returns:
+            [`~pipelines.z_image.ZImagePipelineOutput`] or `tuple`: [`~pipelines.z_image.ZImagePipelineOutput`] if
+            `return_dict` is True, otherwise a `tuple`. When returning a tuple, the first element is a list with the
+            generated images.
+        """
+        # 1. Check inputs and validate strength
+        if strength < 0 or strength > 1:
+            raise ValueError(f"The value of strength should be in [0.0, 1.0] but is {strength}")
+
+        # 2. Preprocess image
+        init_image = self.image_processor.preprocess(image)
+        init_image = init_image.to(dtype=torch.float32)
+
+        # Get dimensions from the preprocessed image if not specified
+        if height is None:
+            height = init_image.shape[-2]
+        if width is None:
+            width = init_image.shape[-1]
+
+        vae_scale = self.vae_scale_factor * 2
+        if height % vae_scale != 0:
+            raise ValueError(
+                f"Height must be divisible by {vae_scale} (got {height}). "
+                f"Please adjust the height to a multiple of {vae_scale}."
+            )
+        if width % vae_scale != 0:
+            raise ValueError(
+                f"Width must be divisible by {vae_scale} (got {width}). "
+                f"Please adjust the width to a multiple of {vae_scale}."
+            )
+
+        device = self._execution_device
+
+        self._guidance_scale = guidance_scale
+        self._joint_attention_kwargs = joint_attention_kwargs
+        self._interrupt = False
+        self._cfg_normalization = cfg_normalization
+        self._cfg_truncation = cfg_truncation
+
+        # 3. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = len(prompt_embeds)
+
+        # If prompt_embeds is provided and prompt is None, skip encoding
+        if prompt_embeds is not None and prompt is None:
+            if self.do_classifier_free_guidance and negative_prompt_embeds is None:
+                raise ValueError(
+                    "When `prompt_embeds` is provided without `prompt`, "
+                    "`negative_prompt_embeds` must also be provided for classifier-free guidance."
+                )
+        else:
+            (
+                prompt_embeds,
+                negative_prompt_embeds,
+            ) = self.encode_prompt(
+                prompt=prompt,
+                negative_prompt=negative_prompt,
+                do_classifier_free_guidance=self.do_classifier_free_guidance,
+                prompt_embeds=prompt_embeds,
+                negative_prompt_embeds=negative_prompt_embeds,
+                device=device,
+                max_sequence_length=max_sequence_length,
+            )
+
+        # 4. Prepare latent variables
+        num_channels_latents = self.transformer.in_channels
+
+        # Repeat prompt_embeds for num_images_per_prompt
+        if num_images_per_prompt > 1:
+            prompt_embeds = [pe for pe in prompt_embeds for _ in range(num_images_per_prompt)]
+            if self.do_classifier_free_guidance and negative_prompt_embeds:
+                negative_prompt_embeds = [npe for npe in negative_prompt_embeds for _ in range(num_images_per_prompt)]
+
+        actual_batch_size = batch_size * num_images_per_prompt
+
+        # Calculate latent dimensions for image_seq_len
+        latent_height = 2 * (int(height) // (self.vae_scale_factor * 2))
+        latent_width = 2 * (int(width) // (self.vae_scale_factor * 2))
+        image_seq_len = (latent_height // 2) * (latent_width // 2)
+
+        # 5. Prepare timesteps
+        mu = calculate_shift(
+            image_seq_len,
+            self.scheduler.config.get("base_image_seq_len", 256),
+            self.scheduler.config.get("max_image_seq_len", 4096),
+            self.scheduler.config.get("base_shift", 0.5),
+            self.scheduler.config.get("max_shift", 1.15),
+        )
+        self.scheduler.sigma_min = 0.0
+        scheduler_kwargs = {"mu": mu}
+        timesteps, num_inference_steps = retrieve_timesteps(
+            self.scheduler,
+            num_inference_steps,
+            device,
+            sigmas=sigmas,
+            **scheduler_kwargs,
+        )
+
+        # 6. Adjust timesteps based on strength
+        timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
+        if num_inference_steps < 1:
+            raise ValueError(
+                f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline "
+                f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline."
+            )
+        latent_timestep = timesteps[:1].repeat(actual_batch_size)
+
+        # 7. Prepare latents from image
+        latents, noise, original_image_latents, latent_image_ids = self.prepare_latents(
+            init_image,
+            latent_timestep,
+            actual_batch_size,
+            num_channels_latents,
+            height,
+            width,
+            prompt_embeds[0].dtype,
+            device,
+            generator,
+            latents,
+        )
+        resize_mode = "default"
+        crops_coords = None
+
+        # start diff diff preparation
+        original_mask = self.mask_processor.preprocess(
+            mask_image, height=height, width=width, resize_mode=resize_mode, crops_coords=crops_coords
+        )
+
+        masked_image = init_image * original_mask
+        original_mask, _ = self.prepare_mask_latents(
+            original_mask,
+            masked_image,
+            batch_size,
+            num_images_per_prompt,
+            height,
+            width,
+            prompt_embeds[0].dtype,
+            device,
+            generator,
+        )
+        mask_thresholds = torch.arange(num_inference_steps, dtype=original_mask.dtype) / num_inference_steps
+        mask_thresholds = mask_thresholds.reshape(-1, 1, 1, 1).to(device)
+        masks = original_mask > mask_thresholds
+        # end diff diff preparation
+
+        num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+        self._num_timesteps = len(timesteps)
+
+        # 8. Denoising loop
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                if self.interrupt:
+                    continue
+
+                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+                timestep = t.expand(latents.shape[0])
+                timestep = (1000 - timestep) / 1000
+                # Normalized time for time-aware config (0 at start, 1 at end)
+                t_norm = timestep[0].item()
+
+                # Handle cfg truncation
+                current_guidance_scale = self.guidance_scale
+                if (
+                    self.do_classifier_free_guidance
+                    and self._cfg_truncation is not None
+                    and float(self._cfg_truncation) <= 1
+                ):
+                    if t_norm > self._cfg_truncation:
+                        current_guidance_scale = 0.0
+
+                # Run CFG only if configured AND scale is non-zero
+                apply_cfg = self.do_classifier_free_guidance and current_guidance_scale > 0
+
+                if apply_cfg:
+                    latents_typed = latents.to(self.transformer.dtype)
+                    latent_model_input = latents_typed.repeat(2, 1, 1, 1)
+                    prompt_embeds_model_input = prompt_embeds + negative_prompt_embeds
+                    timestep_model_input = timestep.repeat(2)
+                else:
+                    latent_model_input = latents.to(self.transformer.dtype)
+                    prompt_embeds_model_input = prompt_embeds
+                    timestep_model_input = timestep
+
+                latent_model_input = latent_model_input.unsqueeze(2)
+                latent_model_input_list = list(latent_model_input.unbind(dim=0))
+
+                model_out_list = self.transformer(
+                    latent_model_input_list,
+                    timestep_model_input,
+                    prompt_embeds_model_input,
+                )[0]
+
+                if apply_cfg:
+                    # Perform CFG
+                    pos_out = model_out_list[:actual_batch_size]
+                    neg_out = model_out_list[actual_batch_size:]
+
+                    noise_pred = []
+                    for j in range(actual_batch_size):
+                        pos = pos_out[j].float()
+                        neg = neg_out[j].float()
+
+                        pred = pos + current_guidance_scale * (pos - neg)
+
+                        # Renormalization
+                        if self._cfg_normalization and float(self._cfg_normalization) > 0.0:
+                            ori_pos_norm = torch.linalg.vector_norm(pos)
+                            new_pos_norm = torch.linalg.vector_norm(pred)
+                            max_new_norm = ori_pos_norm * float(self._cfg_normalization)
+                            if new_pos_norm > max_new_norm:
+                                pred = pred * (max_new_norm / new_pos_norm)
+
+                        noise_pred.append(pred)
+
+                    noise_pred = torch.stack(noise_pred, dim=0)
+                else:
+                    noise_pred = torch.stack([t.float() for t in model_out_list], dim=0)
+
+                noise_pred = noise_pred.squeeze(2)
+                noise_pred = -noise_pred
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents = self.scheduler.step(noise_pred.to(torch.float32), t, latents, return_dict=False)[0]
+                assert latents.dtype == torch.float32
+
+                # start diff diff
+                image_latent = original_image_latents
+                latents_dtype = latents.dtype
+                if i < len(timesteps) - 1:
+                    noise_timestep = timesteps[i + 1]
+                    image_latent = self.scheduler.scale_noise(
+                        original_image_latents, torch.tensor([noise_timestep]), noise
+                    )
+
+                    mask = masks[i].to(latents_dtype)
+                    latents = image_latent * mask + latents * (1 - mask)
+                # end diff diff
+
+                if latents.dtype != latents_dtype:
+                    if torch.backends.mps.is_available():
+                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                        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)
+
+                # 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.to(self.vae.dtype)
+            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)
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (image,)
+
+        return ZImagePipelineOutput(images=image)

setup.py

@@ -274,7 +274,7 @@ version_range_max = max(sys.version_info[1], 10) + 1
 
 setup(
     name="diffusers",
-    version="0.36.0.dev0",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
+    version="0.37.0.dev0",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
     description="State-of-the-art diffusion in PyTorch and JAX.",
     long_description=open("README.md", "r", encoding="utf-8").read(),
     long_description_content_type="text/markdown",

src/diffusers/loaders/single_file_model.py

@@ -162,7 +162,7 @@ SINGLE_FILE_LOADABLE_CLASSES = {
         "default_subfolder": "transformer",
     },
     "QwenImageTransformer2DModel": {
-        "checkpoint_mapping_fn": lambda x: x,
+        "checkpoint_mapping_fn": lambda checkpoint, **kwargs: checkpoint,
         "default_subfolder": "transformer",
     },
     "Flux2Transformer2DModel": {

src/diffusers/loaders/single_file_utils.py

@@ -120,7 +120,10 @@ CHECKPOINT_KEY_NAMES = {
     "hunyuan-video": "txt_in.individual_token_refiner.blocks.0.adaLN_modulation.1.bias",
     "instruct-pix2pix": "model.diffusion_model.input_blocks.0.0.weight",
     "lumina2": ["model.diffusion_model.cap_embedder.0.weight", "cap_embedder.0.weight"],
-    "z-image-turbo": "cap_embedder.0.weight",
+    "z-image-turbo": [
+        "model.diffusion_model.layers.0.adaLN_modulation.0.weight",
+        "layers.0.adaLN_modulation.0.weight",
+    ],
     "z-image-turbo-controlnet": "control_all_x_embedder.2-1.weight",
     "z-image-turbo-controlnet-2.x": "control_layers.14.adaLN_modulation.0.weight",
     "sana": [
@@ -223,7 +226,8 @@ DIFFUSERS_DEFAULT_PIPELINE_PATHS = {
     "cosmos-2.0-v2w-14B": {"pretrained_model_name_or_path": "nvidia/Cosmos-Predict2-14B-Video2World"},
     "z-image-turbo": {"pretrained_model_name_or_path": "Tongyi-MAI/Z-Image-Turbo"},
     "z-image-turbo-controlnet": {"pretrained_model_name_or_path": "hlky/Z-Image-Turbo-Fun-Controlnet-Union"},
-    "z-image-turbo-controlnet-2.x": {"pretrained_model_name_or_path": "hlky/Z-Image-Turbo-Fun-Controlnet-Union-2.1"},
+    "z-image-turbo-controlnet-2.0": {"pretrained_model_name_or_path": "hlky/Z-Image-Turbo-Fun-Controlnet-Union-2.0"},
+    "z-image-turbo-controlnet-2.1": {"pretrained_model_name_or_path": "hlky/Z-Image-Turbo-Fun-Controlnet-Union-2.1"},
 }
 
 # Use to configure model sample size when original config is provided
@@ -727,10 +731,7 @@ def infer_diffusers_model_type(checkpoint):
     ):
         model_type = "instruct-pix2pix"
 
-    elif (
-        CHECKPOINT_KEY_NAMES["z-image-turbo"] in checkpoint
-        and checkpoint[CHECKPOINT_KEY_NAMES["z-image-turbo"]].shape[0] == 2560
-    ):
+    elif any(key in checkpoint for key in CHECKPOINT_KEY_NAMES["z-image-turbo"]):
         model_type = "z-image-turbo"
 
     elif any(key in checkpoint for key in CHECKPOINT_KEY_NAMES["lumina2"]):
@@ -784,7 +785,13 @@ def infer_diffusers_model_type(checkpoint):
             raise ValueError(f"Unexpected x_embedder shape: {x_embedder_shape} when loading Cosmos 2.0 model.")
 
     elif CHECKPOINT_KEY_NAMES["z-image-turbo-controlnet-2.x"] in checkpoint:
-        model_type = "z-image-turbo-controlnet-2.x"
+        before_proj_weight = checkpoint.get("control_noise_refiner.0.before_proj.weight", None)
+        if before_proj_weight is None:
+            model_type = "z-image-turbo-controlnet-2.0"
+        elif before_proj_weight is not None and torch.all(before_proj_weight == 0.0):
+            model_type = "z-image-turbo-controlnet-2.0"
+        else:
+            model_type = "z-image-turbo-controlnet-2.1"
 
     elif CHECKPOINT_KEY_NAMES["z-image-turbo-controlnet"] in checkpoint:
         model_type = "z-image-turbo-controlnet"
@@ -3852,6 +3859,7 @@ def convert_z_image_transformer_checkpoint_to_diffusers(checkpoint, **kwargs):
         ".attention.k_norm.weight": ".attention.norm_k.weight",
         ".attention.q_norm.weight": ".attention.norm_q.weight",
         ".attention.out.weight": ".attention.to_out.0.weight",
+        "model.diffusion_model.": "",
     }
 
     def convert_z_image_fused_attention(key: str, state_dict: dict[str, object]) -> None:
@@ -3886,6 +3894,9 @@ def convert_z_image_transformer_checkpoint_to_diffusers(checkpoint, **kwargs):
 
         update_state_dict(converted_state_dict, key, new_key)
 
+    if "norm_final.weight" in converted_state_dict.keys():
+        _ = converted_state_dict.pop("norm_final.weight")
+
     # Handle any special logic which can't be expressed by a simple 1:1 remapping with the handlers in
     # special_keys_remap
     for key in list(converted_state_dict.keys()):

src/diffusers/models/autoencoders/autoencoder_kl_hunyuanimage.py

@@ -27,7 +27,7 @@ from ...utils.accelerate_utils import apply_forward_hook
 from ..activations import get_activation
 from ..modeling_outputs import AutoencoderKLOutput
 from ..modeling_utils import ModelMixin
-from .vae import DecoderOutput, DiagonalGaussianDistribution
+from .vae import AutoencoderMixin, DecoderOutput, DiagonalGaussianDistribution
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -410,7 +410,7 @@ class HunyuanImageDecoder2D(nn.Module):
         return h
 
 
-class AutoencoderKLHunyuanImage(ModelMixin, ConfigMixin, FromOriginalModelMixin):
+class AutoencoderKLHunyuanImage(ModelMixin, AutoencoderMixin, ConfigMixin, FromOriginalModelMixin):
     r"""
     A VAE model for 2D images with spatial tiling support.
 
@@ -486,27 +486,6 @@ class AutoencoderKLHunyuanImage(ModelMixin, ConfigMixin, FromOriginalModelMixin)
         self.tile_overlap_factor = tile_overlap_factor or self.tile_overlap_factor
         self.tile_latent_min_size = self.tile_sample_min_size // self.config.spatial_compression_ratio
 
-    def disable_tiling(self) -> None:
-        r"""
-        Disable tiled VAE decoding. If `enable_tiling` was previously enabled, this method will go back to computing
-        decoding in one step.
-        """
-        self.use_tiling = False
-
-    def enable_slicing(self) -> None:
-        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.use_slicing = True
-
-    def disable_slicing(self) -> None:
-        r"""
-        Disable sliced VAE decoding. If `enable_slicing` was previously enabled, this method will go back to computing
-        decoding in one step.
-        """
-        self.use_slicing = False
-
     def _encode(self, x: torch.Tensor):
 
         batch_size, num_channels, height, width = x.shape

src/diffusers/models/autoencoders/autoencoder_kl_hunyuanimage_refiner.py

@@ -26,7 +26,7 @@ from ...utils.accelerate_utils import apply_forward_hook
 from ..activations import get_activation
 from ..modeling_outputs import AutoencoderKLOutput
 from ..modeling_utils import ModelMixin
-from .vae import DecoderOutput, DiagonalGaussianDistribution
+from .vae import AutoencoderMixin, DecoderOutput, DiagonalGaussianDistribution
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -584,7 +584,7 @@ class HunyuanImageRefinerDecoder3D(nn.Module):
         return hidden_states
 
 
-class AutoencoderKLHunyuanImageRefiner(ModelMixin, ConfigMixin):
+class AutoencoderKLHunyuanImageRefiner(ModelMixin, AutoencoderMixin, ConfigMixin):
     r"""
     A VAE model with KL loss for encoding videos into latents and decoding latent representations into videos. Used for
     HunyuanImage-2.1 Refiner.
@@ -685,27 +685,6 @@ class AutoencoderKLHunyuanImageRefiner(ModelMixin, ConfigMixin):
         self.tile_sample_stride_width = tile_sample_stride_width or self.tile_sample_stride_width
         self.tile_overlap_factor = tile_overlap_factor or self.tile_overlap_factor
 
-    def disable_tiling(self) -> None:
-        r"""
-        Disable tiled VAE decoding. If `enable_tiling` was previously enabled, this method will go back to computing
-        decoding in one step.
-        """
-        self.use_tiling = False
-
-    def enable_slicing(self) -> None:
-        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.use_slicing = True
-
-    def disable_slicing(self) -> None:
-        r"""
-        Disable sliced VAE decoding. If `enable_slicing` was previously enabled, this method will go back to computing
-        decoding in one step.
-        """
-        self.use_slicing = False
-
     def _encode(self, x: torch.Tensor) -> torch.Tensor:
         _, _, _, height, width = x.shape
 

src/diffusers/models/autoencoders/autoencoder_kl_hunyuanvideo15.py

@@ -26,7 +26,7 @@ from ...utils.accelerate_utils import apply_forward_hook
 from ..activations import get_activation
 from ..modeling_outputs import AutoencoderKLOutput
 from ..modeling_utils import ModelMixin
-from .vae import DecoderOutput, DiagonalGaussianDistribution
+from .vae import AutoencoderMixin, DecoderOutput, DiagonalGaussianDistribution
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -625,7 +625,7 @@ class HunyuanVideo15Decoder3D(nn.Module):
         return hidden_states
 
 
-class AutoencoderKLHunyuanVideo15(ModelMixin, ConfigMixin):
+class AutoencoderKLHunyuanVideo15(ModelMixin, AutoencoderMixin, ConfigMixin):
     r"""
     A VAE model with KL loss for encoding videos into latents and decoding latent representations into videos. Used for
     HunyuanVideo-1.5.
@@ -723,27 +723,6 @@ class AutoencoderKLHunyuanVideo15(ModelMixin, ConfigMixin):
         self.tile_latent_min_width = tile_latent_min_width or self.tile_latent_min_width
         self.tile_overlap_factor = tile_overlap_factor or self.tile_overlap_factor
 
-    def disable_tiling(self) -> None:
-        r"""
-        Disable tiled VAE decoding. If `enable_tiling` was previously enabled, this method will go back to computing
-        decoding in one step.
-        """
-        self.use_tiling = False
-
-    def enable_slicing(self) -> None:
-        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.use_slicing = True
-
-    def disable_slicing(self) -> None:
-        r"""
-        Disable sliced VAE decoding. If `enable_slicing` was previously enabled, this method will go back to computing
-        decoding in one step.
-        """
-        self.use_slicing = False
-
     def _encode(self, x: torch.Tensor) -> torch.Tensor:
         _, _, _, height, width = x.shape
 

src/diffusers/models/transformers/transformer_wan.py

@@ -134,7 +134,8 @@ class WanAttnProcessor:
                 dropout_p=0.0,
                 is_causal=False,
                 backend=self._attention_backend,
-                parallel_config=self._parallel_config,
+                # Reference: https://github.com/huggingface/diffusers/pull/12909
+                parallel_config=None,
             )
             hidden_states_img = hidden_states_img.flatten(2, 3)
             hidden_states_img = hidden_states_img.type_as(query)
@@ -147,7 +148,8 @@ class WanAttnProcessor:
             dropout_p=0.0,
             is_causal=False,
             backend=self._attention_backend,
-            parallel_config=self._parallel_config,
+            # Reference: https://github.com/huggingface/diffusers/pull/12909
+            parallel_config=(self._parallel_config if encoder_hidden_states is None else None),
         )
         hidden_states = hidden_states.flatten(2, 3)
         hidden_states = hidden_states.type_as(query)
@@ -552,9 +554,11 @@ class WanTransformer3DModel(
         "blocks.0": {
             "hidden_states": ContextParallelInput(split_dim=1, expected_dims=3, split_output=False),
         },
-        "blocks.*": {
-            "encoder_hidden_states": ContextParallelInput(split_dim=1, expected_dims=3, split_output=False),
-        },
+        # Reference: https://github.com/huggingface/diffusers/pull/12909
+        # We need to disable the splitting of encoder_hidden_states because the image_encoder
+        # (Wan 2.1 I2V) consistently generates 257 tokens for image_embed. This causes the shape
+        # of encoder_hidden_states—whose token count is always 769 (512 + 257) after concatenation
+        # —to be indivisible by the number of devices in the CP.
         "proj_out": ContextParallelOutput(gather_dim=1, expected_dims=3),
         "": {
             "timestep": ContextParallelInput(split_dim=1, expected_dims=2, split_output=False),

src/diffusers/models/transformers/transformer_wan_animate.py

@@ -609,7 +609,8 @@ class WanAttnProcessor:
                 dropout_p=0.0,
                 is_causal=False,
                 backend=self._attention_backend,
-                parallel_config=self._parallel_config,
+                # Reference: https://github.com/huggingface/diffusers/pull/12909
+                parallel_config=None,
             )
             hidden_states_img = hidden_states_img.flatten(2, 3)
             hidden_states_img = hidden_states_img.type_as(query)
@@ -622,7 +623,8 @@ class WanAttnProcessor:
             dropout_p=0.0,
             is_causal=False,
             backend=self._attention_backend,
-            parallel_config=self._parallel_config,
+            # Reference: https://github.com/huggingface/diffusers/pull/12909
+            parallel_config=(self._parallel_config if encoder_hidden_states is None else None),
         )
         hidden_states = hidden_states.flatten(2, 3)
         hidden_states = hidden_states.type_as(query)

src/diffusers/pipelines/cosmos/pipeline_cosmos2_5_predict.py

@@ -76,7 +76,7 @@ EXAMPLE_DOC_STRING = """
 
         >>> model_id = "nvidia/Cosmos-Predict2.5-2B"
         >>> pipe = Cosmos2_5_PredictBasePipeline.from_pretrained(
-        ...     model_id, revision="diffusers/base/pre-trianed", torch_dtype=torch.bfloat16
+        ...     model_id, revision="diffusers/base/post-trained", torch_dtype=torch.bfloat16
         ... )
         >>> pipe = pipe.to("cuda")
 

src/diffusers/pipelines/z_image/pipeline_z_image_controlnet.py

@@ -58,14 +58,13 @@ EXAMPLE_DOC_STRING = """
         >>> #     torch_dtype=torch.bfloat16,
         >>> # )
 
-        >>> # 2.0 - `config` is required
+        >>> # 2.0
         >>> # controlnet = ZImageControlNetModel.from_single_file(
         >>> #     hf_hub_download(
         >>> #         "alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.0",
         >>> #         filename="Z-Image-Turbo-Fun-Controlnet-Union-2.0.safetensors",
         >>> #     ),
         >>> #     torch_dtype=torch.bfloat16,
-        >>> #     config="hlky/Z-Image-Turbo-Fun-Controlnet-Union-2.0",
         >>> # )
 
         >>> pipe = ZImageControlNetPipeline.from_pretrained(

src/diffusers/pipelines/z_image/pipeline_z_image_controlnet_inpaint.py

@@ -50,14 +50,13 @@ EXAMPLE_DOC_STRING = """
         ...     torch_dtype=torch.bfloat16,
         ... )
 
-        >>> # 2.0 - `config` is required
+        >>> # 2.0
         >>> # controlnet = ZImageControlNetModel.from_single_file(
         >>> #     hf_hub_download(
         >>> #         "alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.0",
         >>> #         filename="Z-Image-Turbo-Fun-Controlnet-Union-2.0.safetensors",
         >>> #     ),
         >>> #     torch_dtype=torch.bfloat16,
-        >>> #     config="hlky/Z-Image-Turbo-Fun-Controlnet-Union-2.0",
         >>> # )
 
         >>> pipe = ZImageControlNetInpaintPipeline.from_pretrained(

src/diffusers/quantizers/torchao/torchao_quantizer.py

@@ -36,6 +36,9 @@ from ...utils import (
 from ..base import DiffusersQuantizer
 
 
+logger = logging.get_logger(__name__)
+
+
 if TYPE_CHECKING:
     from ...models.modeling_utils import ModelMixin
 
@@ -83,11 +86,19 @@ def _update_torch_safe_globals():
     ]
     try:
         from torchao.dtypes import NF4Tensor
-        from torchao.dtypes.floatx.float8_layout import Float8AQTTensorImpl
-        from torchao.dtypes.uintx.uint4_layout import UInt4Tensor
         from torchao.dtypes.uintx.uintx_layout import UintxAQTTensorImpl, UintxTensor
 
-        safe_globals.extend([UintxTensor, UInt4Tensor, UintxAQTTensorImpl, Float8AQTTensorImpl, NF4Tensor])
+        safe_globals.extend([UintxTensor, UintxAQTTensorImpl, NF4Tensor])
+
+        # note: is_torchao_version(">=", "0.16.0") does not work correctly
+        # with torchao nightly, so using a ">" check which does work correctly
+        if is_torchao_version(">", "0.15.0"):
+            pass
+        else:
+            from torchao.dtypes.floatx.float8_layout import Float8AQTTensorImpl
+            from torchao.dtypes.uintx.uint4_layout import UInt4Tensor
+
+            safe_globals.extend([UInt4Tensor, Float8AQTTensorImpl])
 
     except (ImportError, ModuleNotFoundError) as e:
         logger.warning(
@@ -123,9 +134,6 @@ def fuzzy_match_size(config_name: str) -> Optional[str]:
     return None
 
 
-logger = logging.get_logger(__name__)
-
-
 def _quantization_type(weight):
     from torchao.dtypes import AffineQuantizedTensor
     from torchao.quantization.linear_activation_quantized_tensor import LinearActivationQuantizedTensor