diffusers 0.29.2__py3-none-any.whl → 0.30.0__py3-none-any.whl

diffusers/pipelines/animatediff/pipeline_animatediff_video2video.py

@@ -19,7 +19,7 @@ import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
 
 from ...image_processor import PipelineImageInput
-from ...loaders import IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel, UNetMotionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...models.unets.unet_motion_model import MotionAdapter
@@ -35,6 +35,7 @@ from ...utils import USE_PEFT_BACKEND, logging, scale_lora_layers, unscale_lora_
 from ...utils.torch_utils import randn_tensor
 from ...video_processor import VideoProcessor
 from ..free_init_utils import FreeInitMixin
+from ..free_noise_utils import AnimateDiffFreeNoiseMixin
 from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
 from .pipeline_output import AnimateDiffPipelineOutput
 
@@ -174,8 +175,9 @@ class AnimateDiffVideoToVideoPipeline(
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
     IPAdapterMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     FreeInitMixin,
+    AnimateDiffFreeNoiseMixin,
 ):
     r"""
     Pipeline for video-to-video generation.
@@ -185,8 +187,8 @@ class AnimateDiffVideoToVideoPipeline(
 
     The pipeline also inherits the following loading methods:
         - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
-        - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights
-        - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
         - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
 
     Args:
@@ -288,7 +290,7 @@ class AnimateDiffVideoToVideoPipeline(
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -421,7 +423,7 @@ class AnimateDiffVideoToVideoPipeline(
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 
@@ -456,6 +458,9 @@ class AnimateDiffVideoToVideoPipeline(
     def prepare_ip_adapter_image_embeds(
         self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance
     ):
+        image_embeds = []
+        if do_classifier_free_guidance:
+            negative_image_embeds = []
         if ip_adapter_image_embeds is None:
             if not isinstance(ip_adapter_image, list):
                 ip_adapter_image = [ip_adapter_image]
@@ -465,7 +470,6 @@ class AnimateDiffVideoToVideoPipeline(
                     f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
                 )
 
-            image_embeds = []
             for single_ip_adapter_image, image_proj_layer in zip(
                 ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
             ):
@@ -473,46 +477,52 @@ class AnimateDiffVideoToVideoPipeline(
                 single_image_embeds, single_negative_image_embeds = self.encode_image(
                     single_ip_adapter_image, device, 1, output_hidden_state
                 )
-                single_image_embeds = torch.stack([single_image_embeds] * num_images_per_prompt, dim=0)
-                single_negative_image_embeds = torch.stack(
-                    [single_negative_image_embeds] * num_images_per_prompt, dim=0
-                )
 
+                image_embeds.append(single_image_embeds[None, :])
                 if do_classifier_free_guidance:
-                    single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds])
-                    single_image_embeds = single_image_embeds.to(device)
-
-                image_embeds.append(single_image_embeds)
+                    negative_image_embeds.append(single_negative_image_embeds[None, :])
         else:
-            repeat_dims = [1]
-            image_embeds = []
             for single_image_embeds in ip_adapter_image_embeds:
                 if do_classifier_free_guidance:
                     single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
-                    single_image_embeds = single_image_embeds.repeat(
-                        num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:]))
-                    )
-                    single_negative_image_embeds = single_negative_image_embeds.repeat(
-                        num_images_per_prompt, *(repeat_dims * len(single_negative_image_embeds.shape[1:]))
-                    )
-                    single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds])
-                else:
-                    single_image_embeds = single_image_embeds.repeat(
-                        num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:]))
-                    )
+                    negative_image_embeds.append(single_negative_image_embeds)
                 image_embeds.append(single_image_embeds)
 
-        return image_embeds
+        ip_adapter_image_embeds = []
+        for i, single_image_embeds in enumerate(image_embeds):
+            single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+            if do_classifier_free_guidance:
+                single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0)
+                single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0)
 
-    # Copied from diffusers.pipelines.text_to_video_synthesis/pipeline_text_to_video_synth.TextToVideoSDPipeline.decode_latents
-    def decode_latents(self, latents):
+            single_image_embeds = single_image_embeds.to(device=device)
+            ip_adapter_image_embeds.append(single_image_embeds)
+
+        return ip_adapter_image_embeds
+
+    def encode_video(self, video, generator, decode_chunk_size: int = 16) -> torch.Tensor:
+        latents = []
+        for i in range(0, len(video), decode_chunk_size):
+            batch_video = video[i : i + decode_chunk_size]
+            batch_video = retrieve_latents(self.vae.encode(batch_video), generator=generator)
+            latents.append(batch_video)
+        return torch.cat(latents)
+
+    # Copied from diffusers.pipelines.animatediff.pipeline_animatediff.AnimateDiffPipeline.decode_latents
+    def decode_latents(self, latents, decode_chunk_size: int = 16):
         latents = 1 / self.vae.config.scaling_factor * latents
 
         batch_size, channels, num_frames, height, width = latents.shape
         latents = latents.permute(0, 2, 1, 3, 4).reshape(batch_size * num_frames, channels, height, width)
 
-        image = self.vae.decode(latents).sample
-        video = image[None, :].reshape((batch_size, num_frames, -1) + image.shape[2:]).permute(0, 2, 1, 3, 4)
+        video = []
+        for i in range(0, latents.shape[0], decode_chunk_size):
+            batch_latents = latents[i : i + decode_chunk_size]
+            batch_latents = self.vae.decode(batch_latents).sample
+            video.append(batch_latents)
+
+        video = torch.cat(video)
+        video = video[None, :].reshape((batch_size, num_frames, -1) + video.shape[2:]).permute(0, 2, 1, 3, 4)
         # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
         video = video.float()
         return video
@@ -628,6 +638,7 @@ class AnimateDiffVideoToVideoPipeline(
         device,
         generator,
         latents=None,
+        decode_chunk_size: int = 16,
     ):
         if latents is None:
             num_frames = video.shape[1]
@@ -662,13 +673,11 @@ class AnimateDiffVideoToVideoPipeline(
                     )
 
                 init_latents = [
-                    retrieve_latents(self.vae.encode(video[i]), generator=generator[i]).unsqueeze(0)
+                    self.encode_video(video[i], generator[i], decode_chunk_size).unsqueeze(0)
                     for i in range(batch_size)
                 ]
             else:
-                init_latents = [
-                    retrieve_latents(self.vae.encode(vid), generator=generator).unsqueeze(0) for vid in video
-                ]
+                init_latents = [self.encode_video(vid, generator, decode_chunk_size).unsqueeze(0) for vid in video]
 
             init_latents = torch.cat(init_latents, dim=0)
 
@@ -753,6 +762,7 @@ class AnimateDiffVideoToVideoPipeline(
         clip_skip: Optional[int] = None,
         callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
         callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        decode_chunk_size: int = 16,
     ):
         r"""
         The call function to the pipeline for generation.
@@ -828,6 +838,8 @@ class AnimateDiffVideoToVideoPipeline(
                 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.
+            decode_chunk_size (`int`, defaults to `16`):
+                The number of frames to decode at a time when calling `decode_latents` method.
 
         Examples:
 
@@ -929,6 +941,7 @@ class AnimateDiffVideoToVideoPipeline(
             device=device,
             generator=generator,
             latents=latents,
+            decode_chunk_size=decode_chunk_size,
         )
 
         # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
@@ -996,7 +1009,7 @@ class AnimateDiffVideoToVideoPipeline(
         if output_type == "latent":
             video = latents
         else:
-            video_tensor = self.decode_latents(latents)
+            video_tensor = self.decode_latents(latents, decode_chunk_size)
             video = self.video_processor.postprocess_video(video=video_tensor, output_type=output_type)
 
         # 10. Offload all models

diffusers/pipelines/audioldm2/modeling_audioldm2.py

@@ -544,7 +544,7 @@ class AudioLDM2UNet2DConditionModel(ModelMixin, ConfigMixin, UNet2DConditionLoad
 
         def fn_recursive_add_processors(name: str, module: torch.nn.Module, processors: Dict[str, AttentionProcessor]):
             if hasattr(module, "get_processor"):
-                processors[f"{name}.processor"] = module.get_processor(return_deprecated_lora=True)
+                processors[f"{name}.processor"] = module.get_processor()
 
             for sub_name, child in module.named_children():
                 fn_recursive_add_processors(f"{name}.{sub_name}", child, processors)

diffusers/pipelines/audioldm2/pipeline_audioldm2.py

@@ -286,6 +286,7 @@ class AudioLDM2Pipeline(DiffusionPipeline):
                 The sequence of generated hidden-states.
         """
         max_new_tokens = max_new_tokens if max_new_tokens is not None else self.language_model.config.max_new_tokens
+        model_kwargs = self.language_model._get_initial_cache_position(inputs_embeds, model_kwargs)
         for _ in range(max_new_tokens):
             # prepare model inputs
             model_inputs = prepare_inputs_for_generation(inputs_embeds, **model_kwargs)

diffusers/pipelines/aura_flow/__init__.py

@@ -0,0 +1,48 @@
+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 = {}
+_import_structure = {}
+
+
+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["pipeline_aura_flow"] = ["AuraFlowPipeline"]
+
+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 *
+    else:
+        from .pipeline_aura_flow import AuraFlowPipeline
+
+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)