Merge branch 'main' into cogvideox-fun/control

sayakpaul · web-flow · commit 756931ffdd54 · 2024-10-16T07:43:02.000+05:30
diff --git a/docs/source/en/_toctree.yml b/docs/source/en/_toctree.yml
@@ -75,6 +75,8 @@
     title: Outpainting
   title: Advanced inference
 - sections:
+  - local: using-diffusers/cogvideox
+    title: CogVideoX
   - local: using-diffusers/sdxl
     title: Stable Diffusion XL
   - local: using-diffusers/sdxl_turbo
@@ -129,6 +131,8 @@
       title: T2I-Adapters
     - local: training/instructpix2pix
       title: InstructPix2Pix
+    - local: training/cogvideox
+      title: CogVideoX
     title: Models
   - isExpanded: false
     sections:
diff --git a/docs/source/en/training/cogvideox.md b/docs/source/en/training/cogvideox.md
diff --git a/docs/source/en/using-diffusers/cogvideox.md b/docs/source/en/using-diffusers/cogvideox.md
@@ -0,0 +1,120 @@
+<!--Copyright 2024 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.
+-->
+# CogVideoX
+
+CogVideoX is a text-to-video generation model focused on creating more coherent videos aligned with a prompt. It achieves this using several methods.
+
+- a 3D variational autoencoder that compresses videos spatially and temporally, improving compression rate and video accuracy.
+
+- an expert transformer block to help align text and video, and a 3D full attention module for capturing and creating spatially and temporally accurate videos.
+ 
+
+
+## Load model checkpoints
+Model weights may be stored in separate subfolders on the Hub or locally, in which case, you should use the [`~DiffusionPipeline.from_pretrained`] method.
+
+
+```py
+from diffusers import CogVideoXPipeline, CogVideoXImageToVideoPipeline
+pipe = CogVideoXPipeline.from_pretrained(
+    "THUDM/CogVideoX-2b",
+    torch_dtype=torch.float16
+)
+
+pipe = CogVideoXImageToVideoPipeline.from_pretrained(
+    "THUDM/CogVideoX-5b-I2V",
+    torch_dtype=torch.bfloat16
+)
+
+```
+
+## Text-to-Video
+For text-to-video, pass a text prompt. By default, CogVideoX generates a 720x480 video for the best results.
+
+```py
+import torch
+from diffusers import CogVideoXPipeline
+from diffusers.utils import export_to_video
+
+prompt = "An elderly gentleman, with a serene expression, sits at the water's edge, a steaming cup of tea by his side. He is engrossed in his artwork, brush in hand, as he renders an oil painting on a canvas that's propped up against a small, weathered table. The sea breeze whispers through his silver hair, gently billowing his loose-fitting white shirt, while the salty air adds an intangible element to his masterpiece in progress. The scene is one of tranquility and inspiration, with the artist's canvas capturing the vibrant hues of the setting sun reflecting off the tranquil sea."
+
+pipe = CogVideoXPipeline.from_pretrained(
+    "THUDM/CogVideoX-5b",
+    torch_dtype=torch.bfloat16
+)
+
+pipe.enable_model_cpu_offload()
+pipe.vae.enable_tiling()
+
+video = pipe(
+    prompt=prompt,
+    num_videos_per_prompt=1,
+    num_inference_steps=50,
+    num_frames=49,
+    guidance_scale=6,
+    generator=torch.Generator(device="cuda").manual_seed(42),
+).frames[0]
+
+export_to_video(video, "output.mp4", fps=8)
+
+```
+
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cogvideox/cogvideox_out.gif" alt="generated image of an astronaut in a jungle"/>
+</div>
+
+
+## Image-to-Video
+
+
+You'll use the [THUDM/CogVideoX-5b-I2V](https://huggingface.co/THUDM/CogVideoX-5b-I2V)  checkpoint for this guide.
+
+```py
+import torch
+from diffusers import CogVideoXImageToVideoPipeline
+from diffusers.utils import export_to_video, load_image
+
+prompt = "A vast, shimmering ocean flows gracefully under a twilight sky, its waves undulating in a mesmerizing dance of blues and greens. The surface glints with the last rays of the setting sun, casting golden highlights that ripple across the water. Seagulls soar above, their cries blending with the gentle roar of the waves. The horizon stretches infinitely, where the ocean meets the sky in a seamless blend of hues. Close-ups reveal the intricate patterns of the waves, capturing the fluidity and dynamic beauty of the sea in motion."
+image = load_image(image="cogvideox_rocket.png")
+pipe = CogVideoXImageToVideoPipeline.from_pretrained(
+    "THUDM/CogVideoX-5b-I2V",
+    torch_dtype=torch.bfloat16
+)
+ 
+pipe.vae.enable_tiling()
+pipe.vae.enable_slicing()
+
+video = pipe(
+    prompt=prompt,
+    image=image,
+    num_videos_per_prompt=1,
+    num_inference_steps=50,
+    num_frames=49,
+    guidance_scale=6,
+    generator=torch.Generator(device="cuda").manual_seed(42),
+).frames[0]
+
+export_to_video(video, "output.mp4", fps=8)
+```
+
+<div class="flex gap-4">
+  <div>
+    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cogvideox/cogvideox_rocket.png"/>
+    <figcaption class="mt-2 text-center text-sm text-gray-500">initial image</figcaption>
+  </div>
+  <div>
+    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cogvideox/cogvideox_outrocket.gif"/>
+    <figcaption class="mt-2 text-center text-sm text-gray-500">generated video</figcaption>
+  </div>
+</div>
+
diff --git a/docs/source/en/using-diffusers/text-img2vid.md b/docs/source/en/using-diffusers/text-img2vid.md
@@ -23,6 +23,59 @@ This guide will show you how to generate videos, how to configure video model pa
 
 [Stable Video Diffusions (SVD)](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid), [I2VGen-XL](https://huggingface.co/ali-vilab/i2vgen-xl/), [AnimateDiff](https://huggingface.co/guoyww/animatediff), and [ModelScopeT2V](https://huggingface.co/ali-vilab/text-to-video-ms-1.7b) are popular models used for video diffusion. Each model is distinct. For example, AnimateDiff inserts a motion modeling module into a frozen text-to-image model to generate personalized animated images, whereas SVD is entirely pretrained from scratch with a three-stage training process to generate short high-quality videos.
 
+[CogVideoX](https://huggingface.co/collections/THUDM/cogvideo-66c08e62f1685a3ade464cce) is another popular video generation model. The model is a multidimensional transformer that integrates text, time, and space. It employs full attention in the attention module and includes an expert block at the layer level to spatially align text and video.
+
+### CogVideoX
+ 
+[CogVideoX](../api/pipelines/cogvideox) uses a 3D Variational Autoencoder (VAE) to compress videos along the spatial and temporal dimensions.
+
+Begin by loading the [`CogVideoXPipeline`] and passing an initial text or image to generate a video.
+<Tip>
+
+CogVideoX is available for image-to-video and text-to-video. [THUDM/CogVideoX-5b-I2V](https://huggingface.co/THUDM/CogVideoX-5b-I2V) uses the [`CogVideoXImageToVideoPipeline`] for image-to-video. [THUDM/CogVideoX-5b](https://huggingface.co/THUDM/CogVideoX-5b) and [THUDM/CogVideoX-2b](https://huggingface.co/THUDM/CogVideoX-2b) are available for text-to-video with the [`CogVideoXPipeline`].
+ 
+</Tip>
+
+```py
+import torch
+from diffusers import CogVideoXImageToVideoPipeline
+from diffusers.utils import export_to_video, load_image
+
+prompt = "A vast, shimmering ocean flows gracefully under a twilight sky, its waves undulating in a mesmerizing dance of blues and greens. The surface glints with the last rays of the setting sun, casting golden highlights that ripple across the water. Seagulls soar above, their cries blending with the gentle roar of the waves. The horizon stretches infinitely, where the ocean meets the sky in a seamless blend of hues. Close-ups reveal the intricate patterns of the waves, capturing the fluidity and dynamic beauty of the sea in motion."
+image = load_image(image="cogvideox_rocket.png")
+pipe = CogVideoXImageToVideoPipeline.from_pretrained(
+    "THUDM/CogVideoX-5b-I2V",
+    torch_dtype=torch.bfloat16
+)
+ 
+pipe.vae.enable_tiling()
+pipe.vae.enable_slicing()
+
+video = pipe(
+    prompt=prompt,
+    image=image,
+    num_videos_per_prompt=1,
+    num_inference_steps=50,
+    num_frames=49,
+    guidance_scale=6,
+    generator=torch.Generator(device="cuda").manual_seed(42),
+).frames[0]
+
+export_to_video(video, "output.mp4", fps=8)
+```
+
+<div class="flex gap-4">
+  <div>
+    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cogvideox/cogvideox_rocket.png"/>
+    <figcaption class="mt-2 text-center text-sm text-gray-500">initial image</figcaption>
+  </div>
+  <div>
+    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cogvideox/cogvideox_outrocket.gif"/>
+    <figcaption class="mt-2 text-center text-sm text-gray-500">generated video</figcaption>
+  </div>
+</div>
+
+ 
 ### Stable Video Diffusion
 
 [SVD](../api/pipelines/svd) is based on the Stable Diffusion 2.1 model and it is trained on images, then low-resolution videos, and finally a smaller dataset of high-resolution videos. This model generates a short 2-4 second video from an initial image. You can learn more details about model, like micro-conditioning, in the [Stable Video Diffusion](../using-diffusers/svd) guide.
diff --git a/src/diffusers/models/controlnet_flux.py b/src/diffusers/models/controlnet_flux.py
@@ -23,7 +23,7 @@
 from ..models.attention_processor import AttentionProcessor
 from ..models.modeling_utils import ModelMixin
 from ..utils import USE_PEFT_BACKEND, is_torch_version, logging, scale_lora_layers, unscale_lora_layers
-from .controlnet import BaseOutput, zero_module
+from .controlnet import BaseOutput, ControlNetConditioningEmbedding, zero_module
 from .embeddings import CombinedTimestepGuidanceTextProjEmbeddings, CombinedTimestepTextProjEmbeddings, FluxPosEmbed
 from .modeling_outputs import Transformer2DModelOutput
 from .transformers.transformer_flux import FluxSingleTransformerBlock, FluxTransformerBlock
@@ -55,6 +55,7 @@ def __init__(
         guidance_embeds: bool = False,
         axes_dims_rope: List[int] = [16, 56, 56],
         num_mode: int = None,
+        conditioning_embedding_channels: int = None,
     ):
         super().__init__()
         self.out_channels = in_channels
@@ -106,7 +107,14 @@ def __init__(
         if self.union:
             self.controlnet_mode_embedder = nn.Embedding(num_mode, self.inner_dim)
 
-        self.controlnet_x_embedder = zero_module(torch.nn.Linear(in_channels, self.inner_dim))
+        if conditioning_embedding_channels is not None:
+            self.input_hint_block = ControlNetConditioningEmbedding(
+                conditioning_embedding_channels=conditioning_embedding_channels, block_out_channels=(16, 16, 16, 16)
+            )
+            self.controlnet_x_embedder = torch.nn.Linear(in_channels, self.inner_dim)
+        else:
+            self.input_hint_block = None
+            self.controlnet_x_embedder = zero_module(torch.nn.Linear(in_channels, self.inner_dim))
 
         self.gradient_checkpointing = False
 
@@ -269,6 +277,16 @@ def forward(
                 )
         hidden_states = self.x_embedder(hidden_states)
 
+        if self.input_hint_block is not None:
+            controlnet_cond = self.input_hint_block(controlnet_cond)
+            batch_size, channels, height_pw, width_pw = controlnet_cond.shape
+            height = height_pw // self.config.patch_size
+            width = width_pw // self.config.patch_size
+            controlnet_cond = controlnet_cond.reshape(
+                batch_size, channels, height, self.config.patch_size, width, self.config.patch_size
+            )
+            controlnet_cond = controlnet_cond.permute(0, 2, 4, 1, 3, 5)
+            controlnet_cond = controlnet_cond.reshape(batch_size, height * width, -1)
         # add
         hidden_states = hidden_states + self.controlnet_x_embedder(controlnet_cond)
 
diff --git a/src/diffusers/models/transformers/transformer_flux.py b/src/diffusers/models/transformers/transformer_flux.py
@@ -402,6 +402,7 @@ def forward(
         controlnet_block_samples=None,
         controlnet_single_block_samples=None,
         return_dict: bool = True,
+        controlnet_blocks_repeat: bool = False,
     ) -> Union[torch.FloatTensor, Transformer2DModelOutput]:
         """
         The [`FluxTransformer2DModel`] forward method.
@@ -508,7 +509,13 @@ def custom_forward(*inputs):
             if controlnet_block_samples is not None:
                 interval_control = len(self.transformer_blocks) / len(controlnet_block_samples)
                 interval_control = int(np.ceil(interval_control))
-                hidden_states = hidden_states + controlnet_block_samples[index_block // interval_control]
+                # For Xlabs ControlNet.
+                if controlnet_blocks_repeat:
+                    hidden_states = (
+                        hidden_states + controlnet_block_samples[index_block % len(controlnet_block_samples)]
+                    )
+                else:
+                    hidden_states = hidden_states + controlnet_block_samples[index_block // interval_control]
 
         hidden_states = torch.cat([encoder_hidden_states, hidden_states], dim=1)
 
diff --git a/src/diffusers/pipelines/flux/pipeline_flux_controlnet.py b/src/diffusers/pipelines/flux/pipeline_flux_controlnet.py
@@ -754,19 +754,22 @@ def __call__(
             )
             height, width = control_image.shape[-2:]
 
-            # vae encode
-            control_image = self.vae.encode(control_image).latent_dist.sample()
-            control_image = (control_image - self.vae.config.shift_factor) * self.vae.config.scaling_factor
-
-            # pack
-            height_control_image, width_control_image = control_image.shape[2:]
-            control_image = self._pack_latents(
-                control_image,
-                batch_size * num_images_per_prompt,
-                num_channels_latents,
-                height_control_image,
-                width_control_image,
-            )
+            # xlab controlnet has a input_hint_block and instantx controlnet does not
+            controlnet_blocks_repeat = False if self.controlnet.input_hint_block is None else True
+            if self.controlnet.input_hint_block is None:
+                # vae encode
+                control_image = self.vae.encode(control_image).latent_dist.sample()
+                control_image = (control_image - self.vae.config.shift_factor) * self.vae.config.scaling_factor
+
+                # pack
+                height_control_image, width_control_image = control_image.shape[2:]
+                control_image = self._pack_latents(
+                    control_image,
+                    batch_size * num_images_per_prompt,
+                    num_channels_latents,
+                    height_control_image,
+                    width_control_image,
+                )
 
             # Here we ensure that `control_mode` has the same length as the control_image.
             if control_mode is not None:
@@ -777,8 +780,9 @@ def __call__(
 
         elif isinstance(self.controlnet, FluxMultiControlNetModel):
             control_images = []
-
-            for control_image_ in control_image:
+            # xlab controlnet has a input_hint_block and instantx controlnet does not
+            controlnet_blocks_repeat = False if self.controlnet.nets[0].input_hint_block is None else True
+            for i, control_image_ in enumerate(control_image):
                 control_image_ = self.prepare_image(
                     image=control_image_,
                     width=width,
@@ -790,20 +794,20 @@ def __call__(
                 )
                 height, width = control_image_.shape[-2:]
 
-                # vae encode
-                control_image_ = self.vae.encode(control_image_).latent_dist.sample()
-                control_image_ = (control_image_ - self.vae.config.shift_factor) * self.vae.config.scaling_factor
-
-                # pack
-                height_control_image, width_control_image = control_image_.shape[2:]
-                control_image_ = self._pack_latents(
-                    control_image_,
-                    batch_size * num_images_per_prompt,
-                    num_channels_latents,
-                    height_control_image,
-                    width_control_image,
-                )
-
+                if self.controlnet.nets[0].input_hint_block is None:
+                    # vae encode
+                    control_image_ = self.vae.encode(control_image_).latent_dist.sample()
+                    control_image_ = (control_image_ - self.vae.config.shift_factor) * self.vae.config.scaling_factor
+
+                    # pack
+                    height_control_image, width_control_image = control_image_.shape[2:]
+                    control_image_ = self._pack_latents(
+                        control_image_,
+                        batch_size * num_images_per_prompt,
+                        num_channels_latents,
+                        height_control_image,
+                        width_control_image,
+                    )
                 control_images.append(control_image_)
 
             control_image = control_images
@@ -927,6 +931,7 @@ def __call__(
                     img_ids=latent_image_ids,
                     joint_attention_kwargs=self.joint_attention_kwargs,
                     return_dict=False,
+                    controlnet_blocks_repeat=controlnet_blocks_repeat,
                 )[0]
 
                 # compute the previous noisy sample x_t -> x_t-1
diff --git a/src/diffusers/training_utils.py b/src/diffusers/training_utils.py
@@ -24,6 +24,9 @@
 if is_transformers_available():
     import transformers
 
+    if transformers.integrations.deepspeed.is_deepspeed_zero3_enabled():
+        import deepspeed
+
 if is_peft_available():
     from peft import set_peft_model_state_dict
 
@@ -442,15 +445,13 @@ def step(self, parameters: Iterable[torch.nn.Parameter]):
         self.cur_decay_value = decay
         one_minus_decay = 1 - decay
 
-        context_manager = contextlib.nullcontext
-        if is_transformers_available() and transformers.integrations.deepspeed.is_deepspeed_zero3_enabled():
-            import deepspeed
+        context_manager = contextlib.nullcontext()
 
         if self.foreach:
             if is_transformers_available() and transformers.integrations.deepspeed.is_deepspeed_zero3_enabled():
                 context_manager = deepspeed.zero.GatheredParameters(parameters, modifier_rank=None)
 
-            with context_manager():
+            with context_manager:
                 params_grad = [param for param in parameters if param.requires_grad]
                 s_params_grad = [
                     s_param for s_param, param in zip(self.shadow_params, parameters) if param.requires_grad
@@ -472,7 +473,7 @@ def step(self, parameters: Iterable[torch.nn.Parameter]):
                 if is_transformers_available() and transformers.integrations.deepspeed.is_deepspeed_zero3_enabled():
                     context_manager = deepspeed.zero.GatheredParameters(param, modifier_rank=None)
 
-                with context_manager():
+                with context_manager:
                     if param.requires_grad:
                         s_param.sub_(one_minus_decay * (s_param - param))
                     else:
