diffusers 0.33.1__py3-none-any.whl → 0.35.0__py3-none-any.whl

diffusers/pipelines/pipeline_utils.py

@@ -47,6 +47,7 @@ from ..configuration_utils import ConfigMixin
 from ..models import AutoencoderKL
 from ..models.attention_processor import FusedAttnProcessor2_0
 from ..models.modeling_utils import _LOW_CPU_MEM_USAGE_DEFAULT, ModelMixin
+from ..quantizers import PipelineQuantizationConfig
 from ..quantizers.bitsandbytes.utils import _check_bnb_status
 from ..schedulers.scheduling_utils import SCHEDULER_CONFIG_NAME
 from ..utils import (
@@ -58,6 +59,7 @@ from ..utils import (
     _is_valid_type,
     is_accelerate_available,
     is_accelerate_version,
+    is_hpu_available,
     is_torch_npu_available,
     is_torch_version,
     is_transformers_version,
@@ -65,7 +67,7 @@ from ..utils import (
     numpy_to_pil,
 )
 from ..utils.hub_utils import _check_legacy_sharding_variant_format, load_or_create_model_card, populate_model_card
-from ..utils.torch_utils import is_compiled_module
+from ..utils.torch_utils import empty_device_cache, get_device, is_compiled_module
 
 
 if is_torch_npu_available():
@@ -86,6 +88,7 @@ from .pipeline_loading_utils import (
     _identify_model_variants,
     _maybe_raise_error_for_incorrect_transformers,
     _maybe_raise_warning_for_inpainting,
+    _maybe_warn_for_wrong_component_in_quant_config,
     _resolve_custom_pipeline_and_cls,
     _unwrap_model,
     _update_init_kwargs_with_connected_pipeline,
@@ -105,7 +108,7 @@ LIBRARIES = []
 for library in LOADABLE_CLASSES:
     LIBRARIES.append(library)
 
-SUPPORTED_DEVICE_MAP = ["balanced"]
+SUPPORTED_DEVICE_MAP = ["balanced"] + [get_device()]
 
 logger = logging.get_logger(__name__)
 
@@ -137,6 +140,43 @@ class AudioPipelineOutput(BaseOutput):
     audios: np.ndarray
 
 
+class DeprecatedPipelineMixin:
+    """
+    A mixin that can be used to mark a pipeline as deprecated.
+
+    Pipelines inheriting from this mixin will raise a warning when instantiated, indicating that they are deprecated
+    and won't receive updates past the specified version. Tests will be skipped for pipelines that inherit from this
+    mixin.
+
+    Example usage:
+    ```python
+    class MyDeprecatedPipeline(DeprecatedPipelineMixin, DiffusionPipeline):
+        _last_supported_version = "0.20.0"
+
+        def __init__(self, *args, **kwargs):
+            super().__init__(*args, **kwargs)
+    ```
+    """
+
+    # Override this in the inheriting class to specify the last version that will support this pipeline
+    _last_supported_version = None
+
+    def __init__(self, *args, **kwargs):
+        # Get the class name for the warning message
+        class_name = self.__class__.__name__
+
+        # Get the last supported version or use the current version if not specified
+        version_info = getattr(self.__class__, "_last_supported_version", __version__)
+
+        # Raise a warning that this pipeline is deprecated
+        logger.warning(
+            f"The {class_name} has been deprecated and will not receive bug fixes or feature updates after Diffusers version {version_info}. "
+        )
+
+        # Call the parent class's __init__ method
+        super().__init__(*args, **kwargs)
+
+
 class DiffusionPipeline(ConfigMixin, PushToHubMixin):
     r"""
     Base class for all pipelines.
@@ -404,6 +444,11 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
             if not is_accelerate_available() or is_accelerate_version("<", "0.14.0"):
                 return False
 
+            _, _, is_loaded_in_8bit_bnb = _check_bnb_status(module)
+
+            if is_loaded_in_8bit_bnb:
+                return False
+
             return hasattr(module, "_hf_hook") and (
                 isinstance(module._hf_hook, accelerate.hooks.AlignDevicesHook)
                 or hasattr(module._hf_hook, "hooks")
@@ -445,6 +490,20 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                 f"It seems like you have activated model offloading by calling `enable_model_cpu_offload`, but are now manually moving the pipeline to GPU. It is strongly recommended against doing so as memory gains from offloading are likely to be lost. Offloading automatically takes care of moving the individual components {', '.join(self.components.keys())} to GPU when needed. To make sure offloading works as expected, you should consider moving the pipeline back to CPU: `pipeline.to('cpu')` or removing the move altogether if you use offloading."
             )
 
+        # Enable generic support for Intel Gaudi accelerator using GPU/HPU migration
+        if device_type == "hpu" and kwargs.pop("hpu_migration", True) and is_hpu_available():
+            os.environ["PT_HPU_GPU_MIGRATION"] = "1"
+            logger.debug("Environment variable set: PT_HPU_GPU_MIGRATION=1")
+
+            import habana_frameworks.torch  # noqa: F401
+
+            # HPU hardware check
+            if not (hasattr(torch, "hpu") and torch.hpu.is_available()):
+                raise ValueError("You are trying to call `.to('hpu')` but HPU device is unavailable.")
+
+            os.environ["PT_HPU_MAX_COMPOUND_OP_SIZE"] = "1"
+            logger.debug("Environment variable set: PT_HPU_MAX_COMPOUND_OP_SIZE=1")
+
         module_names, _ = self._get_signature_keys(self)
         modules = [getattr(self, n, None) for n in module_names]
         modules = [m for m in modules if isinstance(m, torch.nn.Module)]
@@ -552,12 +611,12 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                       saved using
                     [`~DiffusionPipeline.save_pretrained`].
                     - A path to a *directory* (for example `./my_pipeline_directory/`) containing a dduf file
-            torch_dtype (`str` or `torch.dtype` or `dict[str, Union[str, torch.dtype]]`, *optional*):
-                Override the default `torch.dtype` and load the model with another dtype. If "auto" is passed, the
-                dtype is automatically derived from the model's weights. To load submodels with different dtype pass a
-                `dict` (for example `{'transformer': torch.bfloat16, 'vae': torch.float16}`). Set the default dtype for
-                unspecified components with `default` (for example `{'transformer': torch.bfloat16, 'default':
-                torch.float16}`). If a component is not specified and no default is set, `torch.float32` is used.
+            torch_dtype (`torch.dtype` or `dict[str, Union[str, torch.dtype]]`, *optional*):
+                Override the default `torch.dtype` and load the model with another dtype. To load submodels with
+                different dtype pass a `dict` (for example `{'transformer': torch.bfloat16, 'vae': torch.float16}`).
+                Set the default dtype for unspecified components with `default` (for example `{'transformer':
+                torch.bfloat16, 'default': torch.float16}`). If a component is not specified and no default is set,
+                `torch.float32` is used.
             custom_pipeline (`str`, *optional*):
 
                 <Tip warning={true}>
@@ -611,14 +670,11 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                 Mirror source to resolve accessibility issues if you’re downloading a model in China. We do not
                 guarantee the timeliness or safety of the source, and you should refer to the mirror site for more
                 information.
-            device_map (`str` or `Dict[str, Union[int, str, torch.device]]`, *optional*):
-                A map that specifies where each submodule should go. It doesn’t need to be defined for each
-                parameter/buffer name; once a given module name is inside, every submodule of it will be sent to the
-                same device.
-
-                Set `device_map="auto"` to have 🤗 Accelerate automatically compute the most optimized `device_map`. For
-                more information about each option see [designing a device
-                map](https://hf.co/docs/accelerate/main/en/usage_guides/big_modeling#designing-a-device-map).
+            device_map (`str`, *optional*):
+                Strategy that dictates how the different components of a pipeline should be placed on available
+                devices. Currently, only "balanced" `device_map` is supported. Check out
+                [this](https://huggingface.co/docs/diffusers/main/en/tutorials/inference_with_big_models#device-placement)
+                to know more.
             max_memory (`Dict`, *optional*):
                 A dictionary device identifier for the maximum memory. Will default to the maximum memory available for
                 each GPU and the available CPU RAM if unset.
@@ -654,8 +710,8 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
 
         <Tip>
 
-        To use private or [gated](https://huggingface.co/docs/hub/models-gated#gated-models) models, log-in with
-        `huggingface-cli login`.
+        To use private or [gated](https://huggingface.co/docs/hub/models-gated#gated-models) models, log-in with `hf
+        auth login`.
 
         </Tip>
 
@@ -705,6 +761,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         use_safetensors = kwargs.pop("use_safetensors", None)
         use_onnx = kwargs.pop("use_onnx", None)
         load_connected_pipeline = kwargs.pop("load_connected_pipeline", False)
+        quantization_config = kwargs.pop("quantization_config", None)
 
         if torch_dtype is not None and not isinstance(torch_dtype, dict) and not isinstance(torch_dtype, torch.dtype):
             torch_dtype = torch.float32
@@ -721,6 +778,9 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                 " install accelerate\n```\n."
             )
 
+        if quantization_config is not None and not isinstance(quantization_config, PipelineQuantizationConfig):
+            raise ValueError("`quantization_config` must be an instance of `PipelineQuantizationConfig`.")
+
         if low_cpu_mem_usage is True and not is_torch_version(">=", "1.9.0"):
             raise NotImplementedError(
                 "Low memory initialization requires torch >= 1.9.0. Please either update your PyTorch version or set"
@@ -925,14 +985,18 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
 
         # 7. Load each module in the pipeline
         current_device_map = None
+        _maybe_warn_for_wrong_component_in_quant_config(init_dict, quantization_config)
         for name, (library_name, class_name) in logging.tqdm(init_dict.items(), desc="Loading pipeline components..."):
             # 7.1 device_map shenanigans
-            if final_device_map is not None and len(final_device_map) > 0:
-                component_device = final_device_map.get(name, None)
-                if component_device is not None:
-                    current_device_map = {"": component_device}
-                else:
-                    current_device_map = None
+            if final_device_map is not None:
+                if isinstance(final_device_map, dict) and len(final_device_map) > 0:
+                    component_device = final_device_map.get(name, None)
+                    if component_device is not None:
+                        current_device_map = {"": component_device}
+                    else:
+                        current_device_map = None
+                elif isinstance(final_device_map, str):
+                    current_device_map = final_device_map
 
             # 7.2 - now that JAX/Flax is an official framework of the library, we might load from Flax names
             class_name = class_name[4:] if class_name.startswith("Flax") else class_name
@@ -981,6 +1045,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                     use_safetensors=use_safetensors,
                     dduf_entries=dduf_entries,
                     provider_options=provider_options,
+                    quantization_config=quantization_config,
                 )
                 logger.info(
                     f"Loaded {name} as {class_name} from `{name}` subfolder of {pretrained_model_name_or_path}."
@@ -1034,6 +1099,8 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         model.register_to_config(_name_or_path=pretrained_model_name_or_path)
         if device_map is not None:
             setattr(model, "hf_device_map", final_device_map)
+        if quantization_config is not None:
+            setattr(model, "quantization_config", quantization_config)
         return model
 
     @property
@@ -1084,19 +1151,20 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                 accelerate.hooks.remove_hook_from_module(model, recurse=True)
         self._all_hooks = []
 
-    def enable_model_cpu_offload(self, gpu_id: Optional[int] = None, device: Union[torch.device, str] = "cuda"):
+    def enable_model_cpu_offload(self, gpu_id: Optional[int] = None, device: Union[torch.device, str] = None):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
-        to `enable_sequential_cpu_offload`, this method moves one whole model at a time to the GPU when its `forward`
-        method is called, and the model remains in GPU until the next model runs. Memory savings are lower than with
-        `enable_sequential_cpu_offload`, but performance is much better due to the iterative execution of the `unet`.
+        to `enable_sequential_cpu_offload`, this method moves one whole model at a time to the accelerator when its
+        `forward` method is called, and the model remains in accelerator until the next model runs. Memory savings are
+        lower than with `enable_sequential_cpu_offload`, but performance is much better due to the iterative execution
+        of the `unet`.
 
         Arguments:
             gpu_id (`int`, *optional*):
                 The ID of the accelerator that shall be used in inference. If not specified, it will default to 0.
-            device (`torch.Device` or `str`, *optional*, defaults to "cuda"):
+            device (`torch.Device` or `str`, *optional*, defaults to None):
                 The PyTorch device type of the accelerator that shall be used in inference. If not specified, it will
-                default to "cuda".
+                automatically detect the available accelerator and use.
         """
         self._maybe_raise_error_if_group_offload_active(raise_error=True)
 
@@ -1118,6 +1186,11 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
 
         self.remove_all_hooks()
 
+        if device is None:
+            device = get_device()
+            if device == "cpu":
+                raise RuntimeError("`enable_model_cpu_offload` requires accelerator, but not found")
+
         torch_device = torch.device(device)
         device_index = torch_device.index
 
@@ -1135,9 +1208,7 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         self._offload_device = device
 
         self.to("cpu", silence_dtype_warnings=True)
-        device_mod = getattr(torch, device.type, None)
-        if hasattr(device_mod, "empty_cache") and device_mod.is_available():
-            device_mod.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
+        empty_device_cache(device.type)
 
         all_model_components = {k: v for k, v in self.components.items() if isinstance(v, torch.nn.Module)}
 
@@ -1196,20 +1267,20 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         # make sure the model is in the same state as before calling it
         self.enable_model_cpu_offload(device=getattr(self, "_offload_device", "cuda"))
 
-    def enable_sequential_cpu_offload(self, gpu_id: Optional[int] = None, device: Union[torch.device, str] = "cuda"):
+    def enable_sequential_cpu_offload(self, gpu_id: Optional[int] = None, device: Union[torch.device, str] = None):
         r"""
         Offloads all models to CPU using 🤗 Accelerate, significantly reducing memory usage. When called, the state
         dicts of all `torch.nn.Module` components (except those in `self._exclude_from_cpu_offload`) are saved to CPU
-        and then moved to `torch.device('meta')` and loaded to GPU only when their specific submodule has its `forward`
-        method called. Offloading happens on a submodule basis. Memory savings are higher than with
+        and then moved to `torch.device('meta')` and loaded to accelerator only when their specific submodule has its
+        `forward` method called. Offloading happens on a submodule basis. Memory savings are higher than with
         `enable_model_cpu_offload`, but performance is lower.
 
         Arguments:
             gpu_id (`int`, *optional*):
                 The ID of the accelerator that shall be used in inference. If not specified, it will default to 0.
-            device (`torch.Device` or `str`, *optional*, defaults to "cuda"):
+            device (`torch.Device` or `str`, *optional*, defaults to None):
                 The PyTorch device type of the accelerator that shall be used in inference. If not specified, it will
-                default to "cuda".
+                automatically detect the available accelerator and use.
         """
         self._maybe_raise_error_if_group_offload_active(raise_error=True)
 
@@ -1225,6 +1296,11 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                 "It seems like you have activated a device mapping strategy on the pipeline so calling `enable_sequential_cpu_offload() isn't allowed. You can call `reset_device_map()` first and then call `enable_sequential_cpu_offload()`."
             )
 
+        if device is None:
+            device = get_device()
+            if device == "cpu":
+                raise RuntimeError("`enable_sequential_cpu_offload` requires accelerator, but not found")
+
         torch_device = torch.device(device)
         device_index = torch_device.index
 
@@ -1242,10 +1318,9 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
         self._offload_device = device
 
         if self.device.type != "cpu":
+            orig_device_type = self.device.type
             self.to("cpu", silence_dtype_warnings=True)
-            device_mod = getattr(torch, self.device.type, None)
-            if hasattr(device_mod, "empty_cache") and device_mod.is_available():
-                device_mod.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
+            empty_device_cache(orig_device_type)
 
         for name, model in self.components.items():
             if not isinstance(model, torch.nn.Module):
@@ -1358,8 +1433,8 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
 
         <Tip>
 
-        To use private or [gated models](https://huggingface.co/docs/hub/models-gated#gated-models), log-in with
-        `huggingface-cli login`.
+        To use private or [gated models](https://huggingface.co/docs/hub/models-gated#gated-models), log-in with `hf
+        auth login
 
         </Tip>
 
@@ -1628,6 +1703,8 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
                 signature_types[k] = (v.annotation,)
             elif get_origin(v.annotation) == Union:
                 signature_types[k] = get_args(v.annotation)
+            elif get_origin(v.annotation) in [List, Dict, list, dict]:
+                signature_types[k] = (v.annotation,)
             else:
                 logger.warning(f"cannot get type annotation for Parameter {k} of {cls}.")
         return signature_types
@@ -1914,11 +1991,13 @@ class DiffusionPipeline(ConfigMixin, PushToHubMixin):
             f"{'' if k.startswith('_') else '_'}{k}": v for k, v in original_config.items() if k not in pipeline_kwargs
         }
 
+        optional_components = (
+            pipeline._optional_components
+            if hasattr(pipeline, "_optional_components") and pipeline._optional_components
+            else []
+        )
         missing_modules = (
-            set(expected_modules)
-            - set(pipeline._optional_components)
-            - set(pipeline_kwargs.keys())
-            - set(true_optional_modules)
+            set(expected_modules) - set(optional_components) - set(pipeline_kwargs.keys()) - set(true_optional_modules)
         )
 
         if len(missing_modules) > 0:
@@ -1990,7 +2069,7 @@ class StableDiffusionMixin:
         self.vae.disable_tiling()
 
     def enable_freeu(self, s1: float, s2: float, b1: float, b2: float):
-        r"""Enables the FreeU mechanism as in https://arxiv.org/abs/2309.11497.
+        r"""Enables the FreeU mechanism as in https://huggingface.co/papers/2309.11497.
 
         The suffixes after the scaling factors represent the stages where they are being applied.
 

diffusers/pipelines/pixart_alpha/pipeline_pixart_alpha.py

@@ -1,4 +1,4 @@
-# Copyright 2024 PixArt-Alpha Authors and The HuggingFace Team. All rights reserved.
+# Copyright 2025 PixArt-Alpha Authors 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.
@@ -256,7 +256,9 @@ class PixArtAlphaPipeline(DiffusionPipeline):
             Tokenizer of class
             [T5Tokenizer](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5Tokenizer).
         transformer ([`PixArtTransformer2DModel`]):
-            A text conditioned `PixArtTransformer2DModel` to denoise the encoded image latents.
+            A text conditioned `PixArtTransformer2DModel` to denoise the encoded image latents. Initially published as
+            [`Transformer2DModel`](https://huggingface.co/PixArt-alpha/PixArt-XL-2-1024-MS/blob/main/transformer/config.json#L2)
+            in the config, but the mismatch can be ignored.
         scheduler ([`SchedulerMixin`]):
             A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
     """
@@ -437,7 +439,7 @@ class PixArtAlphaPipeline(DiffusionPipeline):
     def prepare_extra_step_kwargs(self, generator, eta):
         # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
         # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
-        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
         # and should be between [0, 1]
 
         accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
@@ -598,7 +600,7 @@ class PixArtAlphaPipeline(DiffusionPipeline):
         # &amp
         caption = re.sub(r"&amp", "", caption)
 
-        # ip adresses:
+        # ip addresses:
         caption = re.sub(r"\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3}", " ", caption)
 
         # article ids:
@@ -733,11 +735,11 @@ class PixArtAlphaPipeline(DiffusionPipeline):
                 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 4.5):
-                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.
+                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.
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             height (`int`, *optional*, defaults to self.unet.config.sample_size):
@@ -745,8 +747,8 @@ class PixArtAlphaPipeline(DiffusionPipeline):
             width (`int`, *optional*, defaults to self.unet.config.sample_size):
                 The width in pixels of the generated image.
             eta (`float`, *optional*, defaults to 0.0):
-                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
-                [`schedulers.DDIMScheduler`], will be ignored for others.
+                Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only
+                applies to [`schedulers.DDIMScheduler`], will be ignored for others.
             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.
@@ -832,7 +834,7 @@ class PixArtAlphaPipeline(DiffusionPipeline):
         device = self._execution_device
 
         # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
-        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+        # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
         # corresponds to doing no classifier free guidance.
         do_classifier_free_guidance = guidance_scale > 1.0
 

diffusers/pipelines/pixart_alpha/pipeline_pixart_sigma.py

@@ -1,4 +1,4 @@
-# Copyright 2024 PixArt-Sigma Authors and The HuggingFace Team. All rights reserved.
+# Copyright 2025 PixArt-Sigma Authors 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.
@@ -185,6 +185,26 @@ def retrieve_timesteps(
 class PixArtSigmaPipeline(DiffusionPipeline):
     r"""
     Pipeline for text-to-image generation using PixArt-Sigma.
+
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+    library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+    Args:
+        vae ([`AutoencoderKL`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+        text_encoder ([`T5EncoderModel`]):
+            Frozen text-encoder. PixArt-Alpha uses
+            [T5](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5EncoderModel), specifically the
+            [t5-v1_1-xxl](https://huggingface.co/PixArt-alpha/PixArt-alpha/tree/main/t5-v1_1-xxl) variant.
+        tokenizer (`T5Tokenizer`):
+            Tokenizer of class
+            [T5Tokenizer](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5Tokenizer).
+        transformer ([`PixArtTransformer2DModel`]):
+            A text conditioned `PixArtTransformer2DModel` to denoise the encoded image latents. Initially published as
+            [`Transformer2DModel`](https://huggingface.co/PixArt-alpha/PixArt-Sigma-XL-2-1024-MS/blob/main/transformer/config.json#L2)
+            in the config, but the mismatch can be ignored.
+        scheduler ([`SchedulerMixin`]):
+            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
     """
 
     bad_punct_regex = re.compile(
@@ -363,7 +383,7 @@ class PixArtSigmaPipeline(DiffusionPipeline):
     def prepare_extra_step_kwargs(self, generator, eta):
         # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
         # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
-        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
         # and should be between [0, 1]
 
         accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
@@ -525,7 +545,7 @@ class PixArtSigmaPipeline(DiffusionPipeline):
         # &amp
         caption = re.sub(r"&amp", "", caption)
 
-        # ip adresses:
+        # ip addresses:
         caption = re.sub(r"\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3}", " ", caption)
 
         # article ids:
@@ -660,11 +680,11 @@ class PixArtSigmaPipeline(DiffusionPipeline):
                 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 4.5):
-                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.
+                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.
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             height (`int`, *optional*, defaults to self.unet.config.sample_size):
@@ -672,8 +692,8 @@ class PixArtSigmaPipeline(DiffusionPipeline):
             width (`int`, *optional*, defaults to self.unet.config.sample_size):
                 The width in pixels of the generated image.
             eta (`float`, *optional*, defaults to 0.0):
-                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
-                [`schedulers.DDIMScheduler`], will be ignored for others.
+                Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only
+                applies to [`schedulers.DDIMScheduler`], will be ignored for others.
             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.
@@ -758,7 +778,7 @@ class PixArtSigmaPipeline(DiffusionPipeline):
         device = self._execution_device
 
         # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
-        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+        # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
         # corresponds to doing no classifier free guidance.
         do_classifier_free_guidance = guidance_scale > 1.0
 

diffusers/pipelines/qwenimage/__init__.py

@@ -0,0 +1,55 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+    DIFFUSERS_SLOW_IMPORT,
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    get_objects_from_module,
+    is_torch_available,
+    is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_additional_imports = {}
+_import_structure = {"pipeline_output": ["QwenImagePipelineOutput", "QwenImagePriorReduxPipelineOutput"]}
+
+try:
+    if not (is_transformers_available() and is_torch_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ...utils import dummy_torch_and_transformers_objects  # noqa F403
+
+    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+    _import_structure["modeling_qwenimage"] = ["ReduxImageEncoder"]
+    _import_structure["pipeline_qwenimage"] = ["QwenImagePipeline"]
+    _import_structure["pipeline_qwenimage_edit"] = ["QwenImageEditPipeline"]
+    _import_structure["pipeline_qwenimage_img2img"] = ["QwenImageImg2ImgPipeline"]
+    _import_structure["pipeline_qwenimage_inpaint"] = ["QwenImageInpaintPipeline"]
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+    try:
+        if not (is_transformers_available() and is_torch_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from ...utils.dummy_torch_and_transformers_objects import *  # noqa F403
+    else:
+        from .pipeline_qwenimage import QwenImagePipeline
+        from .pipeline_qwenimage_edit import QwenImageEditPipeline
+        from .pipeline_qwenimage_img2img import QwenImageImg2ImgPipeline
+        from .pipeline_qwenimage_inpaint import QwenImageInpaintPipeline
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(
+        __name__,
+        globals()["__file__"],
+        _import_structure,
+        module_spec=__spec__,
+    )
+
+    for name, value in _dummy_objects.items():
+        setattr(sys.modules[__name__], name, value)
+    for name, value in _additional_imports.items():
+        setattr(sys.modules[__name__], name, value)

diffusers/pipelines/qwenimage/pipeline_output.py

@@ -0,0 +1,21 @@
+from dataclasses import dataclass
+from typing import List, Union
+
+import numpy as np
+import PIL.Image
+
+from ...utils import BaseOutput
+
+
+@dataclass
+class QwenImagePipelineOutput(BaseOutput):
+    """
+    Output class for Stable Diffusion pipelines.
+
+    Args:
+        images (`List[PIL.Image.Image]` or `np.ndarray`)
+            List of denoised PIL images of length `batch_size` or numpy array of shape `(batch_size, height, width,
+            num_channels)`. PIL images or numpy array present the denoised images of the diffusion pipeline.
+    """
+
+    images: Union[List[PIL.Image.Image], np.ndarray]