code release

VITA-Group · Apr 1, 2024 · 8a61e4e · 8a61e4e
1 parent 6a421c1
commit 8a61e4e
Show file tree

Hide file tree

Showing 129 changed files with 25,836 additions and 3 deletions.
diff --git a/README.md b/README.md
@@ -7,6 +7,7 @@ Konstantinos N. Plataniotis, and Zhangyang Wang
 
 ## News
 
+- 2024.4.1:  Released code!
 - 2024.3.25:  Released on arxiv!
 
 ## Overview
@@ -23,13 +24,10 @@ We release our pre-generated static assets in `data/` directory. During training
 
 ## Custom Prompts
 
-## Code is coming soon!
-
 ## Training
 
 ## Testing
 
 ## Citation
 
 If you find this repository/work helpful in your research, please consider citing the paper and starring the repo ⭐.
-
diff --git a/VideoCrafter/License b/VideoCrafter/License
diff --git a/VideoCrafter/README.md b/VideoCrafter/README.md
@@ -0,0 +1,199 @@
+
+## ___***VideoCrafter2: Overcoming Data Limitations for High-Quality Video Diffusion Models***___
+
+<a href='https://ailab-cvc.github.io/videocrafter2/'><img src='https://img.shields.io/badge/Project-Page-green'></a> 
+<a href='https://arxiv.org/abs/2401.09047'><img src='https://img.shields.io/badge/Technique-Report-red'></a> 
+[![Discord](https://dcbadge.vercel.app/api/server/rrayYqZ4tf?style=flat)](https://discord.gg/rrayYqZ4tf)
+<a href='https://huggingface.co/spaces/VideoCrafter/VideoCrafter2'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-blue'></a>
+[![GitHub](https://img.shields.io/github/stars/VideoCrafter/VideoCrafter?style=social)](https://github.com/VideoCrafter/VideoCrafter)
+
+### 🔥🔥 Our dedicated high-resolution I2V model is released at: :point_right:[DynamiCrafter](https://github.com/Doubiiu/DynamiCrafter)!!!
+
+[![](https://img.youtube.com/vi/0NfmIsNAg-g/0.jpg)](https://www.youtube.com/watch?v=0NfmIsNAg-g)
+
+### 🔥The VideoCrafter2 Large improvements over VideoCrafter1 with limited data. Better Motion, Better Concept Combination!!!
+
+Please Join us and create your own film on [Discord/Floor33](https://discord.gg/rrayYqZ4tf).
+
+##### 🎥 Exquisite film, produced by VideoCrafter2, directed by Human
+ [![IMAGE ALT TEXT HERE](https://img.youtube.com/vi/TUsFkW0tK-s/0.jpg)](https://www.youtube.com/watch?v=TUsFkW0tK-s)
+
+## 🔆 Introduction
+
+🤗🤗🤗 VideoCrafter is an open-source video generation and editing toolbox for crafting video content.   
+It currently includes the Text2Video and Image2Video models:
+
+### 1. Generic Text-to-video Generation
+Click the GIF to access the high-resolution video.
+
+<table class="center">
+  <td><a href="https://github.com/AILab-CVC/VideoCrafter/assets/18735168/d20ee09d-fc32-44a8-9e9a-f12f44b30411"><img src=assets/t2v/tom.gif width="320"></td>
+  <td><a href="https://github.com/AILab-CVC/VideoCrafter/assets/18735168/f1d9f434-28e8-44f6-a9b8-cffd67e4574d"><img src=assets/t2v/child.gif width="320"></td>
+  <td><a href="https://github.com/AILab-CVC/VideoCrafter/assets/18735168/bbcfef0e-d8fb-4850-adc0-d8f937c2fa36"><img src=assets/t2v/woman.gif width="320"></td>
+  <tr>
+  <td style="text-align:center;" width="320">"Tom Cruise's face reflects focus, his eyes filled with purpose and drive."</td>
+  <td style="text-align:center;" width="320">"A child excitedly swings on a rusty swing set, laughter filling the air."</td>
+  <td style="text-align:center;" width="320">"A young woman with glasses is jogging in the park wearing a pink headband."</td>
+  <tr>
+</table >
+
+<table class="center">
+  <td><a href="https://github.com/AILab-CVC/VideoCrafter/assets/18735168/7edafc5a-750e-45f3-a46e-b593751a4b12"><img src=assets/t2v/couple.gif width="320"></td>
+  <td><a href="https://github.com/AILab-CVC/VideoCrafter/assets/18735168/37fe41c8-31fb-4e77-bcf9-fa159baa6d86"><img src=assets/t2v/rabbit.gif width="320"></td>
+  <td><a href="https://github.com/AILab-CVC/VideoCrafter/assets/18735168/09791a46-a243-41b8-a6bb-892cdd3a83a2"><img src=assets/t2v/duck.gif width="320"></td>
+  <tr>
+  <td style="text-align:center;" width="320">"With the style of van gogh, A young couple dances under the moonlight by the lake."</td>
+  <td style="text-align:center;" width="320">"A rabbit, low-poly game art style"</td>
+  <td style="text-align:center;" width="320">"Impressionist style, a yellow rubber duck floating on the wave on the sunset"</td>
+  <tr>
+</table >
+
+### 2. Generic Image-to-video Generation
+
+<table class="center">
+  <td><img src=assets/i2v/input/blackswan.png width="170"></td>
+  <td><img src=assets/i2v/input/horse.png width="170"></td>
+  <td><img src=assets/i2v/input/chair.png width="170"></td>
+  <td><img src=assets/i2v/input/sunset.png width="170"></td>
+  <tr>
+  <td><a href="https://github.com/AILab-CVC/VideoCrafter/assets/18735168/1a57edd9-3fd2-4ce9-8313-89aca95b6ec7"><img src=assets/i2v/blackswan.gif width="170"></td>
+  <td><a href="https://github.com/AILab-CVC/VideoCrafter/assets/18735168/d671419d-ae49-4889-807e-b841aef60e8a"><img src=assets/i2v/horse.gif width="170"></td>
+  <td><a href="https://github.com/AILab-CVC/VideoCrafter/assets/18735168/39d730d9-7b47-4132-bdae-4d18f3e651ee"><img src=assets/i2v/chair.gif width="170"></td>
+  <td><a href="https://github.com/AILab-CVC/VideoCrafter/assets/18735168/dc8dd0d5-a80d-4f31-94db-f9ea0b13172b"><img src=assets/i2v/sunset.gif width="170"></td>
+  <tr>
+  <td style="text-align:center;" width="170">"a black swan swims on the pond"</td>
+  <td style="text-align:center;" width="170">"a girl is riding a horse fast on grassland"</td>
+  <td style="text-align:center;" width="170">"a boy sits on a chair facing the sea"</td>
+  <td style="text-align:center;" width="170">"two galleons moving in the wind at sunset"</td>
+
+</table >
+
+:boom: **You are highly recommended to try our dedicated I2V model [DynamiCrafter](https://github.com/Doubiiu/DynamiCrafter): Higher resolution, Better Dynamics, More Coherence!!!**
+
+---
+
+## 📝 Changelog
+- __[2024.02.05]__: 🔥🔥 Release new I2V model with the resolution of 640x1024 of VideoCrafter1/DynamiCrafter. 
+
+- __[2024.01.26]__: Release the 512x320 checkpoint of VideoCrafter2. 
+
+- __[2024.01.18]__: Release the [VideoCrafter2](https://ailab-cvc.github.io/videocrafter2/) and [Tech Report](https://arxiv.org/abs/2401.09047)!
+
+- __[2023.10.30]__: Release [VideoCrafter1](https://arxiv.org/abs/2310.19512) Technical Report!
+
+- __[2023.10.13]__: Release the VideoCrafter1, High Quality Video Generation!
+
+- __[2023.08.14]__: Release a new version of VideoCrafter on [Discord/Floor33](https://discord.gg/uHaQuThT). Please join us to create your own film!
+
+- __[2023.04.18]__: Release a VideoControl model with most of the watermarks removed!
+
+- __[2023.04.05]__: Release pretrained Text-to-Video models, VideoLora models, and inference code.
+<br>
+
+
+## ⏳ Models
+
+|T2V-Models|Resolution|Checkpoints|
+|:---------|:---------|:--------|
+|VideoCrafter2|320x512|[Hugging Face](https://huggingface.co/VideoCrafter/VideoCrafter2/blob/main/model.ckpt)
+|VideoCrafter1|576x1024|[Hugging Face](https://huggingface.co/VideoCrafter/Text2Video-1024/blob/main/model.ckpt)
+|VideoCrafter1|320x512|[Hugging Face](https://huggingface.co/VideoCrafter/Text2Video-512/blob/main/model.ckpt)
+
+|I2V-Models|Resolution|Checkpoints|
+|:---------|:---------|:--------|
+|VideoCrafter1|640x1024|[Hugging Face](https://huggingface.co/Doubiiu/DynamiCrafter_1024/blob/main/model.ckpt)
+|VideoCrafter1|320x512|[Hugging Face](https://huggingface.co/VideoCrafter/Image2Video-512/blob/main/model.ckpt)
+
+
+
+## ⚙️ Setup
+
+### 1. Install Environment via Anaconda (Recommended)
+```bash
+conda create -n videocrafter python=3.8.5
+conda activate videocrafter
+pip install -r requirements.txt
+```
+
+
+## 💫 Inference 
+### 1. Text-to-Video
+
+1) Download pretrained T2V models via [Hugging Face](https://huggingface.co/VideoCrafter/VideoCrafter2/blob/main/model.ckpt), and put the `model.ckpt` in `checkpoints/base_512_v2/model.ckpt`.
+2) Input the following commands in terminal.
+```bash
+  sh scripts/run_text2video.sh
+```
+
+### 2. Image-to-Video
+
+1) Download pretrained I2V models via [Hugging Face](https://huggingface.co/VideoCrafter/Image2Video-512-v1.0/blob/main/model.ckpt), and put the `model.ckpt` in `checkpoints/i2v_512_v1/model.ckpt`.
+2) Input the following commands in terminal.
+```bash
+  sh scripts/run_image2video.sh
+```
+
+### 3. Local Gradio demo
+
+1. Download the pretrained T2V and I2V models and put them in the corresponding directory according to the previous guidelines.
+2. Input the following commands in terminal.
+```bash
+  python gradio_app.py
+```
+
+---
+## 📋 Techinical Report
+😉 VideoCrafter2 Tech report: [VideoCrafter2: Overcoming Data Limitations for High-Quality Video Diffusion Models](https://arxiv.org/abs/2401.09047)
+
+😉 VideoCrafter1 Tech report: [VideoCrafter1: Open Diffusion Models for High-Quality Video Generation](https://arxiv.org/abs/2310.19512)
+<br>
+
+## 😉 Citation
+The technical report is currently unavailable as it is still in preparation. You can cite the paper of our image-to-video model and related base model.
+```
+@misc{chen2024videocrafter2,
+      title={VideoCrafter2: Overcoming Data Limitations for High-Quality Video Diffusion Models}, 
+      author={Haoxin Chen and Yong Zhang and Xiaodong Cun and Menghan Xia and Xintao Wang and Chao Weng and Ying Shan},
+      year={2024},
+      eprint={2401.09047},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV}
+}
+
+@misc{chen2023videocrafter1,
+      title={VideoCrafter1: Open Diffusion Models for High-Quality Video Generation}, 
+      author={Haoxin Chen and Menghan Xia and Yingqing He and Yong Zhang and Xiaodong Cun and Shaoshu Yang and Jinbo Xing and Yaofang Liu and Qifeng Chen and Xintao Wang and Chao Weng and Ying Shan},
+      year={2023},
+      eprint={2310.19512},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV}
+}
+
+@article{xing2023dynamicrafter,
+      title={DynamiCrafter: Animating Open-domain Images with Video Diffusion Priors}, 
+      author={Jinbo Xing and Menghan Xia and Yong Zhang and Haoxin Chen and Xintao Wang and Tien-Tsin Wong and Ying Shan},
+      year={2023},
+      eprint={2310.12190},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV}
+}
+
+@article{he2022lvdm,
+      title={Latent Video Diffusion Models for High-Fidelity Long Video Generation}, 
+      author={Yingqing He and Tianyu Yang and Yong Zhang and Ying Shan and Qifeng Chen},
+      year={2022},
+      eprint={2211.13221},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV}
+}
+```
+
+
+## 🤗 Acknowledgements
+Our codebase builds on [Stable Diffusion](https://github.com/Stability-AI/stablediffusion). 
+Thanks the authors for sharing their awesome codebases! 
+
+
+## 📢 Disclaimer
+We develop this repository for RESEARCH purposes, so it can only be used for personal/research/non-commercial purposes.
+****
diff --git a/VideoCrafter/cog.yaml b/VideoCrafter/cog.yaml
@@ -0,0 +1,25 @@
+# Configuration for Cog ⚙️
+# Reference: https://github.com/replicate/cog/blob/main/docs/yaml.md
+
+build:
+  gpu: true
+  system_packages:
+    - "libgl1-mesa-glx"
+    - "libglib2.0-0"
+  python_version: "3.11"
+  python_packages:
+    - "torch==2.0.1"
+    - "opencv-python==4.8.1.78"
+    - "torchvision==0.15.2"
+    - "pytorch_lightning==2.1.0"
+    - "einops==0.7.0"
+    - "imageio==2.31.6"
+    - "omegaconf==2.3.0"
+    - "transformers==4.35.0"
+    - "moviepy==1.0.3"
+    - "av==10.0.0"
+    - "decord==0.6.0"
+    - "kornia==0.7.0"
+    - "open-clip-torch==2.12.0"
+    - "xformers==0.0.21"
+predict: "predict.py:Predictor"
diff --git a/VideoCrafter/configs/inference_i2v_512_v1.0.yaml b/VideoCrafter/configs/inference_i2v_512_v1.0.yaml
@@ -0,0 +1,83 @@
+model:
+  target: lvdm.models.ddpm3d.LatentVisualDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.012
+    num_timesteps_cond: 1
+    timesteps: 1000
+    first_stage_key: video
+    cond_stage_key: caption
+    cond_stage_trainable: false
+    conditioning_key: crossattn
+    image_size:
+    - 40
+    - 64
+    channels: 4
+    scale_by_std: false
+    scale_factor: 0.18215
+    use_ema: false
+    uncond_type: empty_seq
+    use_scale: true
+    scale_b: 0.7
+    finegrained: true
+    unet_config:
+      target: lvdm.modules.networks.openaimodel3d.UNetModel
+      params:
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions:
+        - 4
+        - 2
+        - 1
+        num_res_blocks: 2
+        channel_mult:
+        - 1
+        - 2
+        - 4
+        - 4
+        num_head_channels: 64
+        transformer_depth: 1
+        context_dim: 1024
+        use_linear: true
+        use_checkpoint: true
+        temporal_conv: true
+        temporal_attention: true
+        temporal_selfatt_only: true
+        use_relative_position: false
+        use_causal_attention: false
+        use_image_attention: true
+        temporal_length: 16
+        addition_attention: true
+        fps_cond: true
+    first_stage_config:
+      target: lvdm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 512
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+    cond_stage_config:
+      target: lvdm.modules.encoders.condition.FrozenOpenCLIPEmbedder
+      params:
+        freeze: true
+        layer: penultimate
+    cond_img_config:
+      target: lvdm.modules.encoders.condition.FrozenOpenCLIPImageEmbedderV2
+      params:
+        freeze: true