diffusers 0.27.0__py3-none-any.whl → 0.32.2__py3-none-any.whl

diffusers/pipelines/animatediff/pipeline_animatediff.py

@@ -15,12 +15,11 @@
 import inspect
 from typing import Any, Callable, Dict, List, Optional, Union
 
-import numpy as np
 import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
 
-from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...image_processor import PipelineImageInput
+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
@@ -41,7 +40,9 @@ from ...utils import (
     unscale_lora_layers,
 )
 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
 
@@ -65,34 +66,14 @@ EXAMPLE_DOC_STRING = """
 """
 
 
-def tensor2vid(video: torch.Tensor, processor: "VaeImageProcessor", output_type: str = "np"):
-    batch_size, channels, num_frames, height, width = video.shape
-    outputs = []
-    for batch_idx in range(batch_size):
-        batch_vid = video[batch_idx].permute(1, 0, 2, 3)
-        batch_output = processor.postprocess(batch_vid, output_type)
-
-        outputs.append(batch_output)
-
-    if output_type == "np":
-        outputs = np.stack(outputs)
-
-    elif output_type == "pt":
-        outputs = torch.stack(outputs)
-
-    elif not output_type == "pil":
-        raise ValueError(f"{output_type} does not exist. Please choose one of ['np', 'pt', 'pil']")
-
-    return outputs
-
-
 class AnimateDiffPipeline(
     DiffusionPipeline,
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
     IPAdapterMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     FreeInitMixin,
+    AnimateDiffFreeNoiseMixin,
 ):
     r"""
     Pipeline for text-to-video generation.
@@ -102,8 +83,8 @@ class AnimateDiffPipeline(
 
     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:
@@ -131,7 +112,7 @@ class AnimateDiffPipeline(
         vae: AutoencoderKL,
         text_encoder: CLIPTextModel,
         tokenizer: CLIPTokenizer,
-        unet: UNet2DConditionModel,
+        unet: Union[UNet2DConditionModel, UNetMotionModel],
         motion_adapter: MotionAdapter,
         scheduler: Union[
             DDIMScheduler,
@@ -159,7 +140,7 @@ class AnimateDiffPipeline(
             image_encoder=image_encoder,
         )
         self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
-        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
+        self.video_processor = VideoProcessor(do_resize=False, vae_scale_factor=self.vae_scale_factor)
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_prompt with num_images_per_prompt -> num_videos_per_prompt
     def encode_prompt(
@@ -169,8 +150,8 @@ class AnimateDiffPipeline(
         num_images_per_prompt,
         do_classifier_free_guidance,
         negative_prompt=None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
         lora_scale: Optional[float] = None,
         clip_skip: Optional[int] = None,
     ):
@@ -190,10 +171,10 @@ class AnimateDiffPipeline(
                 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`).
-            prompt_embeds (`torch.FloatTensor`, *optional*):
+            prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                 provided, text embeddings will be generated from `prompt` input argument.
-            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+            negative_prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                 weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                 argument.
@@ -205,7 +186,7 @@ class AnimateDiffPipeline(
         """
         # 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
@@ -337,9 +318,10 @@ class AnimateDiffPipeline(
             negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
-        if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
-            # Retrieve the original scale by scaling back the LoRA layers
-            unscale_lora_layers(self.text_encoder, lora_scale)
+        if self.text_encoder is not None:
+            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)
 
         return prompt_embeds, negative_prompt_embeds
 
@@ -372,6 +354,9 @@ class AnimateDiffPipeline(
     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]
@@ -381,7 +366,6 @@ class AnimateDiffPipeline(
                     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
             ):
@@ -389,46 +373,43 @@ class AnimateDiffPipeline(
                 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)
+
+            single_image_embeds = single_image_embeds.to(device=device)
+            ip_adapter_image_embeds.append(single_image_embeds)
+
+        return ip_adapter_image_embeds
 
-    # Copied from diffusers.pipelines.text_to_video_synthesis/pipeline_text_to_video_synth.TextToVideoSDPipeline.decode_latents
-    def decode_latents(self, 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
@@ -451,7 +432,6 @@ class AnimateDiffPipeline(
             extra_step_kwargs["generator"] = generator
         return extra_step_kwargs
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.check_inputs
     def check_inputs(
         self,
         prompt,
@@ -489,8 +469,8 @@ class AnimateDiffPipeline(
             raise ValueError(
                 "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
             )
-        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
-            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+        elif prompt is not None and not isinstance(prompt, (str, list, dict)):
+            raise ValueError(f"`prompt` has to be of type `str`, `list` or `dict` but is {type(prompt)=}")
 
         if negative_prompt is not None and negative_prompt_embeds is not None:
             raise ValueError(
@@ -521,10 +501,21 @@ class AnimateDiffPipeline(
                     f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
                 )
 
-    # Copied from diffusers.pipelines.text_to_video_synthesis.pipeline_text_to_video_synth.TextToVideoSDPipeline.prepare_latents
     def prepare_latents(
         self, batch_size, num_channels_latents, num_frames, height, width, dtype, device, generator, latents=None
     ):
+        # If FreeNoise is enabled, generate latents as described in Equation (7) of [FreeNoise](https://arxiv.org/abs/2310.15169)
+        if self.free_noise_enabled:
+            latents = self._prepare_latents_free_noise(
+                batch_size, num_channels_latents, num_frames, height, width, dtype, device, generator, latents
+            )
+
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
         shape = (
             batch_size,
             num_channels_latents,
@@ -532,11 +523,6 @@ class AnimateDiffPipeline(
             height // self.vae_scale_factor,
             width // self.vae_scale_factor,
         )
-        if isinstance(generator, list) and len(generator) != batch_size:
-            raise ValueError(
-                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
-                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
-            )
 
         if latents is None:
             latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
@@ -570,11 +556,15 @@ class AnimateDiffPipeline(
     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,
+        prompt: Optional[Union[str, List[str]]] = None,
         num_frames: Optional[int] = 16,
         height: Optional[int] = None,
         width: Optional[int] = None,
@@ -584,17 +574,18 @@ class AnimateDiffPipeline(
         num_videos_per_prompt: Optional[int] = 1,
         eta: float = 0.0,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        latents: Optional[torch.FloatTensor] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        latents: Optional[torch.Tensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
         ip_adapter_image: Optional[PipelineImageInput] = None,
-        ip_adapter_image_embeds: Optional[List[torch.FloatTensor]] = None,
+        ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
         clip_skip: Optional[int] = None,
         callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
         callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        decode_chunk_size: int = 16,
         **kwargs,
     ):
         r"""
@@ -625,27 +616,26 @@ class AnimateDiffPipeline(
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                 A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                 generation deterministic.
-            latents (`torch.FloatTensor`, *optional*):
+            latents (`torch.Tensor`, *optional*):
                 Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for video
                 generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                 tensor is generated by sampling using the supplied random `generator`. Latents should be of shape
                 `(batch_size, num_channel, num_frames, height, width)`.
-            prompt_embeds (`torch.FloatTensor`, *optional*):
+            prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
                 provided, text embeddings are generated from the `prompt` input argument.
-            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+            negative_prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
                 not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
             ip_adapter_image: (`PipelineImageInput`, *optional*):
                 Optional image input to work with IP Adapters.
-            ip_adapter_image_embeds (`List[torch.FloatTensor]`, *optional*):
-                Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of IP-adapters.
-                Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should contain the negative image embedding
-                if `do_classifier_free_guidance` is set to `True`.
-                If not provided, embeddings are computed from the `ip_adapter_image` input argument.
+            ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+                Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+                IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
+                contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
+                provided, embeddings are computed from the `ip_adapter_image` input argument.
             output_type (`str`, *optional*, defaults to `"pil"`):
-                The output format of the generated video. Choose between `torch.FloatTensor`, `PIL.Image` or
-                `np.array`.
+                The output format of the generated video. Choose between `torch.Tensor`, `PIL.Image` or `np.array`.
             return_dict (`bool`, *optional*, defaults to `True`):
                 Whether or not to return a [`~pipelines.text_to_video_synthesis.TextToVideoSDPipelineOutput`] instead
                 of a plain tuple.
@@ -663,7 +653,9 @@ class AnimateDiffPipeline(
             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 pipeine class.
+                `._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:
 
@@ -712,9 +704,10 @@ class AnimateDiffPipeline(
         self._guidance_scale = guidance_scale
         self._clip_skip = clip_skip
         self._cross_attention_kwargs = cross_attention_kwargs
+        self._interrupt = False
 
         # 2. Define call parameters
-        if prompt is not None and isinstance(prompt, str):
+        if prompt is not None and isinstance(prompt, (str, dict)):
             batch_size = 1
         elif prompt is not None and isinstance(prompt, list):
             batch_size = len(prompt)
@@ -727,22 +720,39 @@ class AnimateDiffPipeline(
         text_encoder_lora_scale = (
             self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
         )
-        prompt_embeds, negative_prompt_embeds = self.encode_prompt(
-            prompt,
-            device,
-            num_videos_per_prompt,
-            self.do_classifier_free_guidance,
-            negative_prompt,
-            prompt_embeds=prompt_embeds,
-            negative_prompt_embeds=negative_prompt_embeds,
-            lora_scale=text_encoder_lora_scale,
-            clip_skip=self.clip_skip,
-        )
-        # For classifier free guidance, we need to do two forward passes.
-        # Here we concatenate the unconditional and text embeddings into a single batch
-        # to avoid doing two forward passes
-        if self.do_classifier_free_guidance:
-            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+        if self.free_noise_enabled:
+            prompt_embeds, negative_prompt_embeds = self._encode_prompt_free_noise(
+                prompt=prompt,
+                num_frames=num_frames,
+                device=device,
+                num_videos_per_prompt=num_videos_per_prompt,
+                do_classifier_free_guidance=self.do_classifier_free_guidance,
+                negative_prompt=negative_prompt,
+                prompt_embeds=prompt_embeds,
+                negative_prompt_embeds=negative_prompt_embeds,
+                lora_scale=text_encoder_lora_scale,
+                clip_skip=self.clip_skip,
+            )
+        else:
+            prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+                prompt,
+                device,
+                num_videos_per_prompt,
+                self.do_classifier_free_guidance,
+                negative_prompt,
+                prompt_embeds=prompt_embeds,
+                negative_prompt_embeds=negative_prompt_embeds,
+                lora_scale=text_encoder_lora_scale,
+                clip_skip=self.clip_skip,
+            )
+
+            # For classifier free guidance, we need to do two forward passes.
+            # Here we concatenate the unconditional and text embeddings into a single batch
+            # to avoid doing two forward passes
+            if self.do_classifier_free_guidance:
+                prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+
+            prompt_embeds = prompt_embeds.repeat_interleave(repeats=num_frames, dim=0)
 
         if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
             image_embeds = self.prepare_ip_adapter_image_embeds(
@@ -792,8 +802,11 @@ class AnimateDiffPipeline(
             num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
 
             # 8. Denoising loop
-            with self.progress_bar(total=num_inference_steps) as progress_bar:
+            with self.progress_bar(total=self._num_timesteps) as progress_bar:
                 for i, t in enumerate(timesteps):
+                    if self.interrupt:
+                        continue
+
                     # expand the latents if we are doing classifier free guidance
                     latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
                     latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
@@ -835,8 +848,8 @@ class AnimateDiffPipeline(
         if output_type == "latent":
             video = latents
         else:
-            video_tensor = self.decode_latents(latents)
-            video = tensor2vid(video_tensor, self.image_processor, output_type=output_type)
+            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
         self.maybe_free_model_hooks()