Commit ·
b7f044b
0
Parent(s):
Duplicate from ali-vilab/i2vgen-xl
Browse filesCo-authored-by: Shiwei Zhang <StevenZhang@users.noreply.huggingface.co>
- .gitattributes +37 -0
- README.MD +283 -0
- README.md +329 -0
- README_diffusers.md +334 -0
- doc/.DS_Store +0 -0
- doc/i2vgen-xl.md +19 -0
- doc/introduction.pdf +3 -0
- feature_extractor/preprocessor_config.json +27 -0
- image_encoder/config.json +23 -0
- image_encoder/model.fp16.safetensors +3 -0
- image_encoder/model.safetensors +3 -0
- model_index.json +33 -0
- models/i2vgen_xl_00854500.pth +3 -0
- models/open_clip_pytorch_model.bin +3 -0
- models/stable_diffusion_image_key_temporal_attention_x1.json +1 -0
- models/v2-1_512-ema-pruned.ckpt +3 -0
- scheduler/scheduler_config.json +19 -0
- source/VGen.jpg +0 -0
- source/fig_vs_vgen.jpg +0 -0
- source/i2vgen_fig_01.jpg +0 -0
- source/i2vgen_fig_02.jpg +0 -0
- source/i2vgen_fig_04.png +3 -0
- source/logo.png +0 -0
- text_encoder/config.json +25 -0
- text_encoder/model.fp16.safetensors +3 -0
- text_encoder/model.safetensors +3 -0
- tokenizer/merges.txt +0 -0
- tokenizer/special_tokens_map.json +30 -0
- tokenizer/tokenizer_config.json +30 -0
- tokenizer/vocab.json +0 -0
- unet/config.json +31 -0
- unet/diffusion_pytorch_model.fp16.safetensors +3 -0
- unet/diffusion_pytorch_model.safetensors +3 -0
- vae/config.json +32 -0
- vae/diffusion_pytorch_model.fp16.safetensors +3 -0
- vae/diffusion_pytorch_model.safetensors +3 -0
.gitattributes
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
| 2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
| 3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
| 4 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
| 5 |
+
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
| 6 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
| 7 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
| 8 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
| 9 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
| 10 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
| 11 |
+
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
| 12 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
| 13 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
| 14 |
+
*.npy filter=lfs diff=lfs merge=lfs -text
|
| 15 |
+
*.npz filter=lfs diff=lfs merge=lfs -text
|
| 16 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
| 17 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
| 18 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
| 19 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
| 20 |
+
*.pickle filter=lfs diff=lfs merge=lfs -text
|
| 21 |
+
*.pkl filter=lfs diff=lfs merge=lfs -text
|
| 22 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
| 23 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
| 24 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
| 25 |
+
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
| 26 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
| 27 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
| 28 |
+
*.tar filter=lfs diff=lfs merge=lfs -text
|
| 29 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
| 30 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
| 31 |
+
*.wasm filter=lfs diff=lfs merge=lfs -text
|
| 32 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
| 33 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
+
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
+
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
doc/introduction.pdf filter=lfs diff=lfs merge=lfs -text
|
| 37 |
+
source/i2vgen_fig_04.png filter=lfs diff=lfs merge=lfs -text
|
README.MD
ADDED
|
@@ -0,0 +1,283 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# VGen
|
| 2 |
+
|
| 3 |
+
|
| 4 |
+
|
| 5 |
+
|
| 6 |
+
VGen is an open-source video synthesis codebase developed by the Tongyi Lab of Alibaba Group, featuring state-of-the-art video generative models. This repository includes implementations of the following methods:
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
- [I2VGen-xl: High-quality image-to-video synthesis via cascaded diffusion models](https://i2vgen-xl.github.io/)
|
| 10 |
+
- [VideoComposer: Compositional Video Synthesis with Motion Controllability](https://videocomposer.github.io/)
|
| 11 |
+
- [Hierarchical Spatio-temporal Decoupling for Text-to-Video Generation](https://higen-t2v.github.io/)
|
| 12 |
+
- [A Recipe for Scaling up Text-to-Video Generation with Text-free Videos]()
|
| 13 |
+
- [InstructVideo: Instructing Video Diffusion Models with Human Feedback]()
|
| 14 |
+
- [DreamVideo: Composing Your Dream Videos with Customized Subject and Motion](https://dreamvideo-t2v.github.io/)
|
| 15 |
+
- [VideoLCM: Video Latent Consistency Model](https://arxiv.org/abs/2312.09109)
|
| 16 |
+
- [Modelscope text-to-video technical report](https://arxiv.org/abs/2308.06571)
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
VGen can produce high-quality videos from the input text, images, desired motion, desired subjects, and even the feedback signals provided. It also offers a variety of commonly used video generation tools such as visualization, sampling, training, inference, join training using images and videos, acceleration, and more.
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
<a href='https://i2vgen-xl.github.io/'><img src='https://img.shields.io/badge/Project-Page-Green'></a> <a href='https://arxiv.org/abs/2311.04145'><img src='https://img.shields.io/badge/Paper-Arxiv-red'></a> [](https://youtu.be/XUi0y7dxqEQ) <a href='https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441039979087.mp4'><img src='source/logo.png'></a>
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
## 🔥News!!!
|
| 26 |
+
- __[2023.12]__ We release the high-efficiency video generation method [VideoLCM](https://arxiv.org/abs/2312.09109)
|
| 27 |
+
- __[2023.12]__ We release the code and model of I2VGen-XL and the ModelScope T2V
|
| 28 |
+
- __[2023.12]__ We release the T2V method [HiGen](https://higen-t2v.github.io) and customizing T2V method [DreamVideo](https://dreamvideo-t2v.github.io).
|
| 29 |
+
- __[2023.12]__ We write an [introduction docment](doc/introduction.pdf) for VGen and compare I2VGen-XL with SVD.
|
| 30 |
+
- __[2023.11]__ We release a high-quality I2VGen-XL model, please refer to the [Webpage](https://i2vgen-xl.github.io)
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
## TODO
|
| 34 |
+
- [x] Release the technical papers and webpage of [I2VGen-XL](doc/i2vgen-xl.md)
|
| 35 |
+
- [x] Release the code and pretrained models that can generate 1280x720 videos
|
| 36 |
+
- [ ] Release models optimized specifically for the human body and faces
|
| 37 |
+
- [ ] Updated version can fully maintain the ID and capture large and accurate motions simultaneously
|
| 38 |
+
- [ ] Release other methods and the corresponding models
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
## Preparation
|
| 42 |
+
|
| 43 |
+
The main features of VGen are as follows:
|
| 44 |
+
- Expandability, allowing for easy management of your own experiments.
|
| 45 |
+
- Completeness, encompassing all common components for video generation.
|
| 46 |
+
- Excellent performance, featuring powerful pre-trained models in multiple tasks.
|
| 47 |
+
|
| 48 |
+
|
| 49 |
+
### Installation
|
| 50 |
+
|
| 51 |
+
```
|
| 52 |
+
conda create -n vgen python=3.8
|
| 53 |
+
conda activate vgen
|
| 54 |
+
pip install torch==1.12.0+cu113 torchvision==0.13.0+cu113 torchaudio==0.12.0 --extra-index-url https://download.pytorch.org/whl/cu113
|
| 55 |
+
pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
|
| 56 |
+
```
|
| 57 |
+
|
| 58 |
+
### Datasets
|
| 59 |
+
|
| 60 |
+
We have provided a **demo dataset** that includes images and videos, along with their lists in ``data``.
|
| 61 |
+
|
| 62 |
+
*Please note that the demo images used here are for testing purposes and were not included in the training.*
|
| 63 |
+
|
| 64 |
+
|
| 65 |
+
### Clone codeb
|
| 66 |
+
|
| 67 |
+
```
|
| 68 |
+
git clone https://github.com/damo-vilab/i2vgen-xl.git
|
| 69 |
+
cd i2vgen-xl
|
| 70 |
+
```
|
| 71 |
+
|
| 72 |
+
|
| 73 |
+
## Getting Started with VGen
|
| 74 |
+
|
| 75 |
+
### (1) Train your text-to-video model
|
| 76 |
+
|
| 77 |
+
|
| 78 |
+
Executing the following command to enable distributed training is as easy as that.
|
| 79 |
+
```
|
| 80 |
+
python train_net.py --cfg configs/t2v_train.yaml
|
| 81 |
+
```
|
| 82 |
+
|
| 83 |
+
In the `t2v_train.yaml` configuration file, you can specify the data, adjust the video-to-image ratio using `frame_lens`, and validate your ideas with different Diffusion settings, and so on.
|
| 84 |
+
|
| 85 |
+
- Before the training, you can download any of our open-source models for initialization. Our codebase supports custom initialization and `grad_scale` settings, all of which are included in the `Pretrain` item in yaml file.
|
| 86 |
+
- During the training, you can view the saved models and intermediate inference results in the `workspace/experiments/t2v_train`directory.
|
| 87 |
+
|
| 88 |
+
After the training is completed, you can perform inference on the model using the following command.
|
| 89 |
+
```
|
| 90 |
+
python inference.py --cfg configs/t2v_infer.yaml
|
| 91 |
+
```
|
| 92 |
+
Then you can find the videos you generated in the `workspace/experiments/test_img_01` directory. For specific configurations such as data, models, seed, etc., please refer to the `t2v_infer.yaml` file.
|
| 93 |
+
|
| 94 |
+
<!-- <table>
|
| 95 |
+
<center>
|
| 96 |
+
<tr>
|
| 97 |
+
<td ><center>
|
| 98 |
+
<video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441754174077.mp4"></video>
|
| 99 |
+
</center></td>
|
| 100 |
+
<td ><center>
|
| 101 |
+
<video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441138824052.mp4"></video>
|
| 102 |
+
</center></td>
|
| 103 |
+
</tr>
|
| 104 |
+
</center>
|
| 105 |
+
</table>
|
| 106 |
+
</center> -->
|
| 107 |
+
|
| 108 |
+
<table>
|
| 109 |
+
<center>
|
| 110 |
+
<tr>
|
| 111 |
+
<td ><center>
|
| 112 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01Ya2I5I25utrJwJ9Jf_!!6000000007587-2-tps-1280-720.png"></image>
|
| 113 |
+
</center></td>
|
| 114 |
+
<td ><center>
|
| 115 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i3/O1CN01CrmYaz1zXBetmg3dd_!!6000000006723-2-tps-1280-720.png"></image>
|
| 116 |
+
</center></td>
|
| 117 |
+
</tr>
|
| 118 |
+
<tr>
|
| 119 |
+
<td ><center>
|
| 120 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441754174077.mp4">HRER</a> to view the generated video.</p>
|
| 121 |
+
</center></td>
|
| 122 |
+
<td ><center>
|
| 123 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441138824052.mp4">HRER</a> to view the generated video.</p>
|
| 124 |
+
</center></td>
|
| 125 |
+
</tr>
|
| 126 |
+
</center>
|
| 127 |
+
</table>
|
| 128 |
+
</center>
|
| 129 |
+
|
| 130 |
+
|
| 131 |
+
### (2) Run the I2VGen-XL model
|
| 132 |
+
|
| 133 |
+
(i) Download model and test data:
|
| 134 |
+
```
|
| 135 |
+
!pip install modelscope
|
| 136 |
+
from modelscope.hub.snapshot_download import snapshot_download
|
| 137 |
+
model_dir = snapshot_download('damo/I2VGen-XL', cache_dir='models/', revision='v1.0.0')
|
| 138 |
+
```
|
| 139 |
+
|
| 140 |
+
(ii) Run the following command:
|
| 141 |
+
```
|
| 142 |
+
python inference.py --cfg configs/i2vgen_xl_infer.yaml
|
| 143 |
+
```
|
| 144 |
+
In a few minutes, you can retrieve the high-definition video you wish to create from the `workspace/experiments/test_img_01` directory. At present, we find that the current model performs inadequately on **anime images** and **images with a black background** due to the lack of relevant training data. We are consistently working to optimize it.
|
| 145 |
+
|
| 146 |
+
|
| 147 |
+
<span style="color:red">Due to the compression of our video quality in GIF format, please click 'HRER' below to view the original video.</span>
|
| 148 |
+
|
| 149 |
+
<center>
|
| 150 |
+
<table>
|
| 151 |
+
<center>
|
| 152 |
+
<tr>
|
| 153 |
+
<td ><center>
|
| 154 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i1/O1CN01CCEq7K1ZeLpNQqrWu_!!6000000003219-0-tps-1280-720.jpg"></image>
|
| 155 |
+
</center></td>
|
| 156 |
+
<td ><center>
|
| 157 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442125067544.mp4"></video> -->
|
| 158 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01hIQcvG1spmQMLqBo0_!!6000000005816-1-tps-1280-704.gif"></image>
|
| 159 |
+
</center></td>
|
| 160 |
+
</tr>
|
| 161 |
+
<tr>
|
| 162 |
+
<td ><center>
|
| 163 |
+
<p>Input Image</p>
|
| 164 |
+
</center></td>
|
| 165 |
+
<td ><center>
|
| 166 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442125067544.mp4">HRER</a> to view the generated video.</p>
|
| 167 |
+
</center></td>
|
| 168 |
+
</tr>
|
| 169 |
+
<tr>
|
| 170 |
+
<td ><center>
|
| 171 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01ZXY7UN23K8q4oQ3uG_!!6000000007236-2-tps-1280-720.png"></image>
|
| 172 |
+
</center></td>
|
| 173 |
+
<td ><center>
|
| 174 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441385957074.mp4"></video> -->
|
| 175 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i1/O1CN01iaSiiv1aJZURUEY53_!!6000000003309-1-tps-1280-704.gif"></image>
|
| 176 |
+
</center></td>
|
| 177 |
+
</tr>
|
| 178 |
+
<tr>
|
| 179 |
+
<td ><center>
|
| 180 |
+
<p>Input Image</p>
|
| 181 |
+
</center></td>
|
| 182 |
+
<td ><center>
|
| 183 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441385957074.mp4">HRER</a> to view the generated video.</p>
|
| 184 |
+
</center></td>
|
| 185 |
+
</tr>
|
| 186 |
+
<tr>
|
| 187 |
+
<td ><center>
|
| 188 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i3/O1CN01NHpVGl1oat4H54Hjf_!!6000000005242-2-tps-1280-720.png"></image>
|
| 189 |
+
</center></td>
|
| 190 |
+
<td ><center>
|
| 191 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442102706767.mp4"></video> -->
|
| 192 |
+
<!-- <image muted="true" height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01DgLj1T240jfpzKoaQ_!!6000000007329-1-tps-1280-704.gif"></image>
|
| 193 |
+
-->
|
| 194 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01DgLj1T240jfpzKoaQ_!!6000000007329-1-tps-1280-704.gif"></image>
|
| 195 |
+
</center></td>
|
| 196 |
+
</tr>
|
| 197 |
+
<tr>
|
| 198 |
+
<td ><center>
|
| 199 |
+
<p>Input Image</p>
|
| 200 |
+
</center></td>
|
| 201 |
+
<td ><center>
|
| 202 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442102706767.mp4">HRER</a> to view the generated video.</p>
|
| 203 |
+
</center></td>
|
| 204 |
+
</tr>
|
| 205 |
+
<tr>
|
| 206 |
+
<td ><center>
|
| 207 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i1/O1CN01odS61s1WW9tXen21S_!!6000000002795-0-tps-1280-720.jpg"></image>
|
| 208 |
+
</center></td>
|
| 209 |
+
<td ><center>
|
| 210 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442163934688.mp4"></video> -->
|
| 211 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i3/O1CN01Jyk1HT28JkZtpAtY6_!!6000000007912-1-tps-1280-704.gif"></image>
|
| 212 |
+
</center></td>
|
| 213 |
+
</tr>
|
| 214 |
+
<tr>
|
| 215 |
+
<td ><center>
|
| 216 |
+
<p>Input Image</p>
|
| 217 |
+
</center></td>
|
| 218 |
+
<td ><center>
|
| 219 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442163934688.mp4">HRER</a> to view the generated video.</p>
|
| 220 |
+
</center></td>
|
| 221 |
+
</tr>
|
| 222 |
+
</center>
|
| 223 |
+
</table>
|
| 224 |
+
</center>
|
| 225 |
+
|
| 226 |
+
### (3) Other methods
|
| 227 |
+
|
| 228 |
+
In preparation.
|
| 229 |
+
|
| 230 |
+
|
| 231 |
+
## Customize your own approach
|
| 232 |
+
|
| 233 |
+
Our codebase essentially supports all the commonly used components in video generation. You can manage your experiments flexibly by adding corresponding registration classes, including `ENGINE, MODEL, DATASETS, EMBEDDER, AUTO_ENCODER, DISTRIBUTION, VISUAL, DIFFUSION, PRETRAIN`, and can be compatible with all our open-source algorithms according to your own needs. If you have any questions, feel free to give us your feedback at any time.
|
| 234 |
+
|
| 235 |
+
|
| 236 |
+
|
| 237 |
+
## BibTeX
|
| 238 |
+
|
| 239 |
+
If this repo is useful to you, please cite our corresponding technical paper.
|
| 240 |
+
|
| 241 |
+
|
| 242 |
+
```bibtex
|
| 243 |
+
@article{2023i2vgenxl,
|
| 244 |
+
title={I2VGen-XL: High-Quality Image-to-Video Synthesis via Cascaded Diffusion Models},
|
| 245 |
+
author={Zhang, Shiwei and Wang, Jiayu and Zhang, Yingya and Zhao, Kang and Yuan, Hangjie and Qing, Zhiwu and Wang, Xiang and Zhao, Deli and Zhou, Jingren},
|
| 246 |
+
booktitle={arXiv preprint arXiv:2311.04145},
|
| 247 |
+
year={2023}
|
| 248 |
+
}
|
| 249 |
+
@article{2023videocomposer,
|
| 250 |
+
title={VideoComposer: Compositional Video Synthesis with Motion Controllability},
|
| 251 |
+
author={Wang, Xiang and Yuan, Hangjie and Zhang, Shiwei and Chen, Dayou and Wang, Jiuniu, and Zhang, Yingya, and Shen, Yujun, and Zhao, Deli and Zhou, Jingren},
|
| 252 |
+
booktitle={arXiv preprint arXiv:2306.02018},
|
| 253 |
+
year={2023}
|
| 254 |
+
}
|
| 255 |
+
@article{wang2023modelscope,
|
| 256 |
+
title={Modelscope text-to-video technical report},
|
| 257 |
+
author={Wang, Jiuniu and Yuan, Hangjie and Chen, Dayou and Zhang, Yingya and Wang, Xiang and Zhang, Shiwei},
|
| 258 |
+
journal={arXiv preprint arXiv:2308.06571},
|
| 259 |
+
year={2023}
|
| 260 |
+
}
|
| 261 |
+
@article{dreamvideo,
|
| 262 |
+
title={DreamVideo: Composing Your Dream Videos with Customized Subject and Motion},
|
| 263 |
+
author={Wei, Yujie and Zhang, Shiwei and Qing, Zhiwu and Yuan, Hangjie and Liu, Zhiheng and Liu, Yu and Zhang, Yingya and Zhou, Jingren and Shan, Hongming},
|
| 264 |
+
journal={arXiv preprint arXiv:2312.04433},
|
| 265 |
+
year={2023}
|
| 266 |
+
}
|
| 267 |
+
@article{qing2023higen,
|
| 268 |
+
title={Hierarchical Spatio-temporal Decoupling for Text-to-Video Generation},
|
| 269 |
+
author={Qing, Zhiwu and Zhang, Shiwei and Wang, Jiayu and Wang, Xiang and Wei, Yujie and Zhang, Yingya and Gao, Changxin and Sang, Nong },
|
| 270 |
+
journal={arXiv preprint arXiv:2312.04483},
|
| 271 |
+
year={2023}
|
| 272 |
+
}
|
| 273 |
+
@article{wang2023videolcm,
|
| 274 |
+
title={VideoLCM: Video Latent Consistency Model},
|
| 275 |
+
author={Wang, Xiang and Zhang, Shiwei and Zhang, Han and Liu, Yu and Zhang, Yingya and Gao, Changxin and Sang, Nong },
|
| 276 |
+
journal={arXiv preprint arXiv:2312.09109},
|
| 277 |
+
year={2023}
|
| 278 |
+
}
|
| 279 |
+
```
|
| 280 |
+
|
| 281 |
+
## Disclaimer
|
| 282 |
+
|
| 283 |
+
This open-source model is trained with using [WebVid-10M](https://m-bain.github.io/webvid-dataset/) and [LAION-400M](https://laion.ai/blog/laion-400-open-dataset/) datasets and is intended for <strong>RESEARCH/NON-COMMERCIAL USE ONLY</strong>.
|
README.md
ADDED
|
@@ -0,0 +1,329 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
tags:
|
| 4 |
+
- image-to-video
|
| 5 |
+
pipeline_tag: text-to-video
|
| 6 |
+
---
|
| 7 |
+
# VGen
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
VGen is an open-source video synthesis codebase developed by the Tongyi Lab of Alibaba Group, featuring state-of-the-art video generative models. This repository includes implementations of the following methods:
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
- [I2VGen-xl: High-quality image-to-video synthesis via cascaded diffusion models](https://i2vgen-xl.github.io/)
|
| 16 |
+
- [VideoComposer: Compositional Video Synthesis with Motion Controllability](https://videocomposer.github.io/)
|
| 17 |
+
- [Hierarchical Spatio-temporal Decoupling for Text-to-Video Generation](https://higen-t2v.github.io/)
|
| 18 |
+
- [A Recipe for Scaling up Text-to-Video Generation with Text-free Videos]()
|
| 19 |
+
- [InstructVideo: Instructing Video Diffusion Models with Human Feedback]()
|
| 20 |
+
- [DreamVideo: Composing Your Dream Videos with Customized Subject and Motion](https://dreamvideo-t2v.github.io/)
|
| 21 |
+
- [VideoLCM: Video Latent Consistency Model](https://arxiv.org/abs/2312.09109)
|
| 22 |
+
- [Modelscope text-to-video technical report](https://arxiv.org/abs/2308.06571)
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
VGen can produce high-quality videos from the input text, images, desired motion, desired subjects, and even the feedback signals provided. It also offers a variety of commonly used video generation tools such as visualization, sampling, training, inference, join training using images and videos, acceleration, and more.
|
| 26 |
+
|
| 27 |
+
|
| 28 |
+
<a href='https://i2vgen-xl.github.io/'><img src='https://img.shields.io/badge/Project-Page-Green'></a> <a href='https://arxiv.org/abs/2311.04145'><img src='https://img.shields.io/badge/Paper-Arxiv-red'></a> [](https://youtu.be/XUi0y7dxqEQ) <a href='https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441039979087.mp4'><img src='source/logo.png'></a>
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
## 🔥News!!!
|
| 32 |
+
- __[2023.12]__ We release the high-efficiency video generation method [VideoLCM](https://arxiv.org/abs/2312.09109)
|
| 33 |
+
- __[2023.12]__ We release the code and model of I2VGen-XL and the ModelScope T2V
|
| 34 |
+
- __[2023.12]__ We release the T2V method [HiGen](https://higen-t2v.github.io) and customizing T2V method [DreamVideo](https://dreamvideo-t2v.github.io).
|
| 35 |
+
- __[2023.12]__ We write an [introduction docment](doc/introduction.pdf) for VGen and compare I2VGen-XL with SVD.
|
| 36 |
+
- __[2023.11]__ We release a high-quality I2VGen-XL model, please refer to the [Webpage](https://i2vgen-xl.github.io)
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
## TODO
|
| 40 |
+
- [x] Release the technical papers and webpage of [I2VGen-XL](doc/i2vgen-xl.md)
|
| 41 |
+
- [x] Release the code and pretrained models that can generate 1280x720 videos
|
| 42 |
+
- [ ] Release models optimized specifically for the human body and faces
|
| 43 |
+
- [ ] Updated version can fully maintain the ID and capture large and accurate motions simultaneously
|
| 44 |
+
- [ ] Release other methods and the corresponding models
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
## Preparation
|
| 48 |
+
|
| 49 |
+
The main features of VGen are as follows:
|
| 50 |
+
- Expandability, allowing for easy management of your own experiments.
|
| 51 |
+
- Completeness, encompassing all common components for video generation.
|
| 52 |
+
- Excellent performance, featuring powerful pre-trained models in multiple tasks.
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
### Installation
|
| 56 |
+
|
| 57 |
+
```
|
| 58 |
+
conda create -n vgen python=3.8
|
| 59 |
+
conda activate vgen
|
| 60 |
+
pip install torch==1.12.0+cu113 torchvision==0.13.0+cu113 torchaudio==0.12.0 --extra-index-url https://download.pytorch.org/whl/cu113
|
| 61 |
+
pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
|
| 62 |
+
```
|
| 63 |
+
|
| 64 |
+
### Datasets
|
| 65 |
+
|
| 66 |
+
We have provided a **demo dataset** that includes images and videos, along with their lists in ``data``.
|
| 67 |
+
|
| 68 |
+
*Please note that the demo images used here are for testing purposes and were not included in the training.*
|
| 69 |
+
|
| 70 |
+
|
| 71 |
+
### Clone codeb
|
| 72 |
+
|
| 73 |
+
```
|
| 74 |
+
git clone https://github.com/damo-vilab/i2vgen-xl.git
|
| 75 |
+
cd i2vgen-xl
|
| 76 |
+
```
|
| 77 |
+
|
| 78 |
+
|
| 79 |
+
## Getting Started with VGen
|
| 80 |
+
|
| 81 |
+
### (1) Train your text-to-video model
|
| 82 |
+
|
| 83 |
+
|
| 84 |
+
Executing the following command to enable distributed training is as easy as that.
|
| 85 |
+
```
|
| 86 |
+
python train_net.py --cfg configs/t2v_train.yaml
|
| 87 |
+
```
|
| 88 |
+
|
| 89 |
+
In the `t2v_train.yaml` configuration file, you can specify the data, adjust the video-to-image ratio using `frame_lens`, and validate your ideas with different Diffusion settings, and so on.
|
| 90 |
+
|
| 91 |
+
- Before the training, you can download any of our open-source models for initialization. Our codebase supports custom initialization and `grad_scale` settings, all of which are included in the `Pretrain` item in yaml file.
|
| 92 |
+
- During the training, you can view the saved models and intermediate inference results in the `workspace/experiments/t2v_train`directory.
|
| 93 |
+
|
| 94 |
+
After the training is completed, you can perform inference on the model using the following command.
|
| 95 |
+
```
|
| 96 |
+
python inference.py --cfg configs/t2v_infer.yaml
|
| 97 |
+
```
|
| 98 |
+
Then you can find the videos you generated in the `workspace/experiments/test_img_01` directory. For specific configurations such as data, models, seed, etc., please refer to the `t2v_infer.yaml` file.
|
| 99 |
+
|
| 100 |
+
<!-- <table>
|
| 101 |
+
<center>
|
| 102 |
+
<tr>
|
| 103 |
+
<td ><center>
|
| 104 |
+
<video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441754174077.mp4"></video>
|
| 105 |
+
</center></td>
|
| 106 |
+
<td ><center>
|
| 107 |
+
<video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441138824052.mp4"></video>
|
| 108 |
+
</center></td>
|
| 109 |
+
</tr>
|
| 110 |
+
</center>
|
| 111 |
+
</table>
|
| 112 |
+
</center> -->
|
| 113 |
+
|
| 114 |
+
<table>
|
| 115 |
+
<center>
|
| 116 |
+
<tr>
|
| 117 |
+
<td ><center>
|
| 118 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01Ya2I5I25utrJwJ9Jf_!!6000000007587-2-tps-1280-720.png"></image>
|
| 119 |
+
</center></td>
|
| 120 |
+
<td ><center>
|
| 121 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i3/O1CN01CrmYaz1zXBetmg3dd_!!6000000006723-2-tps-1280-720.png"></image>
|
| 122 |
+
</center></td>
|
| 123 |
+
</tr>
|
| 124 |
+
<tr>
|
| 125 |
+
<td ><center>
|
| 126 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441754174077.mp4">HRER</a> to view the generated video.</p>
|
| 127 |
+
</center></td>
|
| 128 |
+
<td ><center>
|
| 129 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441138824052.mp4">HRER</a> to view the generated video.</p>
|
| 130 |
+
</center></td>
|
| 131 |
+
</tr>
|
| 132 |
+
</center>
|
| 133 |
+
</table>
|
| 134 |
+
</center>
|
| 135 |
+
|
| 136 |
+
|
| 137 |
+
### (2) Run the I2VGen-XL model
|
| 138 |
+
|
| 139 |
+
(i) Download model and test data:
|
| 140 |
+
```
|
| 141 |
+
!pip install modelscope
|
| 142 |
+
from modelscope.hub.snapshot_download import snapshot_download
|
| 143 |
+
model_dir = snapshot_download('damo/I2VGen-XL', cache_dir='models/', revision='v1.0.0')
|
| 144 |
+
```
|
| 145 |
+
|
| 146 |
+
(ii) Run the following command:
|
| 147 |
+
```
|
| 148 |
+
python inference.py --cfg configs/i2vgen_xl_infer.yaml
|
| 149 |
+
```
|
| 150 |
+
In a few minutes, you can retrieve the high-definition video you wish to create from the `workspace/experiments/test_img_01` directory. At present, we find that the current model performs inadequately on **anime images** and **images with a black background** due to the lack of relevant training data. We are consistently working to optimize it.
|
| 151 |
+
|
| 152 |
+
|
| 153 |
+
<span style="color:red">Due to the compression of our video quality in GIF format, please click 'HRER' below to view the original video.</span>
|
| 154 |
+
|
| 155 |
+
<center>
|
| 156 |
+
<table>
|
| 157 |
+
<center>
|
| 158 |
+
<tr>
|
| 159 |
+
<td ><center>
|
| 160 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i1/O1CN01CCEq7K1ZeLpNQqrWu_!!6000000003219-0-tps-1280-720.jpg"></image>
|
| 161 |
+
</center></td>
|
| 162 |
+
<td ><center>
|
| 163 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442125067544.mp4"></video> -->
|
| 164 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01hIQcvG1spmQMLqBo0_!!6000000005816-1-tps-1280-704.gif"></image>
|
| 165 |
+
</center></td>
|
| 166 |
+
</tr>
|
| 167 |
+
<tr>
|
| 168 |
+
<td ><center>
|
| 169 |
+
<p>Input Image</p>
|
| 170 |
+
</center></td>
|
| 171 |
+
<td ><center>
|
| 172 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442125067544.mp4">HRER</a> to view the generated video.</p>
|
| 173 |
+
</center></td>
|
| 174 |
+
</tr>
|
| 175 |
+
<tr>
|
| 176 |
+
<td ><center>
|
| 177 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01ZXY7UN23K8q4oQ3uG_!!6000000007236-2-tps-1280-720.png"></image>
|
| 178 |
+
</center></td>
|
| 179 |
+
<td ><center>
|
| 180 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441385957074.mp4"></video> -->
|
| 181 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i1/O1CN01iaSiiv1aJZURUEY53_!!6000000003309-1-tps-1280-704.gif"></image>
|
| 182 |
+
</center></td>
|
| 183 |
+
</tr>
|
| 184 |
+
<tr>
|
| 185 |
+
<td ><center>
|
| 186 |
+
<p>Input Image</p>
|
| 187 |
+
</center></td>
|
| 188 |
+
<td ><center>
|
| 189 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441385957074.mp4">HRER</a> to view the generated video.</p>
|
| 190 |
+
</center></td>
|
| 191 |
+
</tr>
|
| 192 |
+
<tr>
|
| 193 |
+
<td ><center>
|
| 194 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i3/O1CN01NHpVGl1oat4H54Hjf_!!6000000005242-2-tps-1280-720.png"></image>
|
| 195 |
+
</center></td>
|
| 196 |
+
<td ><center>
|
| 197 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442102706767.mp4"></video> -->
|
| 198 |
+
<!-- <image muted="true" height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01DgLj1T240jfpzKoaQ_!!6000000007329-1-tps-1280-704.gif"></image>
|
| 199 |
+
-->
|
| 200 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01DgLj1T240jfpzKoaQ_!!6000000007329-1-tps-1280-704.gif"></image>
|
| 201 |
+
</center></td>
|
| 202 |
+
</tr>
|
| 203 |
+
<tr>
|
| 204 |
+
<td ><center>
|
| 205 |
+
<p>Input Image</p>
|
| 206 |
+
</center></td>
|
| 207 |
+
<td ><center>
|
| 208 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442102706767.mp4">HERE</a> to view the generated video.</p>
|
| 209 |
+
</center></td>
|
| 210 |
+
</tr>
|
| 211 |
+
<tr>
|
| 212 |
+
<td ><center>
|
| 213 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i1/O1CN01odS61s1WW9tXen21S_!!6000000002795-0-tps-1280-720.jpg"></image>
|
| 214 |
+
</center></td>
|
| 215 |
+
<td ><center>
|
| 216 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442163934688.mp4"></video> -->
|
| 217 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i3/O1CN01Jyk1HT28JkZtpAtY6_!!6000000007912-1-tps-1280-704.gif"></image>
|
| 218 |
+
</center></td>
|
| 219 |
+
</tr>
|
| 220 |
+
<tr>
|
| 221 |
+
<td ><center>
|
| 222 |
+
<p>Input Image</p>
|
| 223 |
+
</center></td>
|
| 224 |
+
<td ><center>
|
| 225 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442163934688.mp4">HERE</a> to view the generated video.</p>
|
| 226 |
+
</center></td>
|
| 227 |
+
</tr>
|
| 228 |
+
</center>
|
| 229 |
+
</table>
|
| 230 |
+
</center>
|
| 231 |
+
|
| 232 |
+
### (3) Other methods
|
| 233 |
+
|
| 234 |
+
In preparation.
|
| 235 |
+
|
| 236 |
+
|
| 237 |
+
## Customize your own approach
|
| 238 |
+
|
| 239 |
+
Our codebase essentially supports all the commonly used components in video generation. You can manage your experiments flexibly by adding corresponding registration classes, including `ENGINE, MODEL, DATASETS, EMBEDDER, AUTO_ENCODER, DISTRIBUTION, VISUAL, DIFFUSION, PRETRAIN`, and can be compatible with all our open-source algorithms according to your own needs. If you have any questions, feel free to give us your feedback at any time.
|
| 240 |
+
|
| 241 |
+
## Integration of I2VGenXL with 🧨 diffusers
|
| 242 |
+
|
| 243 |
+
I2VGenXL is supported in the 🧨 diffusers library. Here's how to use it:
|
| 244 |
+
|
| 245 |
+
```python
|
| 246 |
+
import torch
|
| 247 |
+
from diffusers import I2VGenXLPipeline
|
| 248 |
+
from diffusers.utils import load_image, export_to_gif
|
| 249 |
+
|
| 250 |
+
repo_id = "ali-vilab/i2vgen-xl"
|
| 251 |
+
pipeline = I2VGenXLPipeline.from_pretrained(repo_id, torch_dtype=torch.float16, variant="fp16").to("cuda")
|
| 252 |
+
|
| 253 |
+
image_url = "https://github.com/ali-vilab/i2vgen-xl/blob/main/data/test_images/img_0009.png?download=true"
|
| 254 |
+
image = load_image(image_url).convert("RGB")
|
| 255 |
+
prompt = "Papers were floating in the air on a table in the library"
|
| 256 |
+
|
| 257 |
+
generator = torch.manual_seed(8888)
|
| 258 |
+
frames = pipeline(
|
| 259 |
+
prompt=prompt,
|
| 260 |
+
image=image,
|
| 261 |
+
generator=generator
|
| 262 |
+
).frames[0]
|
| 263 |
+
|
| 264 |
+
print(export_to_gif(frames))
|
| 265 |
+
```
|
| 266 |
+
|
| 267 |
+
Find the official documentation [here](https://huggingface.co/docs/diffusers/main/en/api/pipelines/i2vgenxl).
|
| 268 |
+
|
| 269 |
+
Sample output with I2VGenXL:
|
| 270 |
+
|
| 271 |
+
<table>
|
| 272 |
+
<tr>
|
| 273 |
+
<td><center>
|
| 274 |
+
masterpiece, bestquality, sunset.
|
| 275 |
+
<br>
|
| 276 |
+
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/i2vgen-xl-example.gif"
|
| 277 |
+
alt="library"
|
| 278 |
+
style="width: 300px;" />
|
| 279 |
+
</center></td>
|
| 280 |
+
</tr>
|
| 281 |
+
</table>
|
| 282 |
+
|
| 283 |
+
## BibTeX
|
| 284 |
+
|
| 285 |
+
If this repo is useful to you, please cite our corresponding technical paper.
|
| 286 |
+
|
| 287 |
+
|
| 288 |
+
```bibtex
|
| 289 |
+
@article{2023i2vgenxl,
|
| 290 |
+
title={I2VGen-XL: High-Quality Image-to-Video Synthesis via Cascaded Diffusion Models},
|
| 291 |
+
author={Zhang, Shiwei and Wang, Jiayu and Zhang, Yingya and Zhao, Kang and Yuan, Hangjie and Qing, Zhiwu and Wang, Xiang and Zhao, Deli and Zhou, Jingren},
|
| 292 |
+
booktitle={arXiv preprint arXiv:2311.04145},
|
| 293 |
+
year={2023}
|
| 294 |
+
}
|
| 295 |
+
@article{2023videocomposer,
|
| 296 |
+
title={VideoComposer: Compositional Video Synthesis with Motion Controllability},
|
| 297 |
+
author={Wang, Xiang and Yuan, Hangjie and Zhang, Shiwei and Chen, Dayou and Wang, Jiuniu, and Zhang, Yingya, and Shen, Yujun, and Zhao, Deli and Zhou, Jingren},
|
| 298 |
+
booktitle={arXiv preprint arXiv:2306.02018},
|
| 299 |
+
year={2023}
|
| 300 |
+
}
|
| 301 |
+
@article{wang2023modelscope,
|
| 302 |
+
title={Modelscope text-to-video technical report},
|
| 303 |
+
author={Wang, Jiuniu and Yuan, Hangjie and Chen, Dayou and Zhang, Yingya and Wang, Xiang and Zhang, Shiwei},
|
| 304 |
+
journal={arXiv preprint arXiv:2308.06571},
|
| 305 |
+
year={2023}
|
| 306 |
+
}
|
| 307 |
+
@article{dreamvideo,
|
| 308 |
+
title={DreamVideo: Composing Your Dream Videos with Customized Subject and Motion},
|
| 309 |
+
author={Wei, Yujie and Zhang, Shiwei and Qing, Zhiwu and Yuan, Hangjie and Liu, Zhiheng and Liu, Yu and Zhang, Yingya and Zhou, Jingren and Shan, Hongming},
|
| 310 |
+
journal={arXiv preprint arXiv:2312.04433},
|
| 311 |
+
year={2023}
|
| 312 |
+
}
|
| 313 |
+
@article{qing2023higen,
|
| 314 |
+
title={Hierarchical Spatio-temporal Decoupling for Text-to-Video Generation},
|
| 315 |
+
author={Qing, Zhiwu and Zhang, Shiwei and Wang, Jiayu and Wang, Xiang and Wei, Yujie and Zhang, Yingya and Gao, Changxin and Sang, Nong },
|
| 316 |
+
journal={arXiv preprint arXiv:2312.04483},
|
| 317 |
+
year={2023}
|
| 318 |
+
}
|
| 319 |
+
@article{wang2023videolcm,
|
| 320 |
+
title={VideoLCM: Video Latent Consistency Model},
|
| 321 |
+
author={Wang, Xiang and Zhang, Shiwei and Zhang, Han and Liu, Yu and Zhang, Yingya and Gao, Changxin and Sang, Nong },
|
| 322 |
+
journal={arXiv preprint arXiv:2312.09109},
|
| 323 |
+
year={2023}
|
| 324 |
+
}
|
| 325 |
+
```
|
| 326 |
+
|
| 327 |
+
## Disclaimer
|
| 328 |
+
|
| 329 |
+
This open-source model is trained with using [WebVid-10M](https://m-bain.github.io/webvid-dataset/) and [LAION-400M](https://laion.ai/blog/laion-400-open-dataset/) datasets and is intended for <strong>RESEARCH/NON-COMMERCIAL USE ONLY</strong>.
|
README_diffusers.md
ADDED
|
@@ -0,0 +1,334 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
library_name: diffusers
|
| 4 |
+
tags:
|
| 5 |
+
- image-to-video
|
| 6 |
+
pipeline_tag: text-to-video
|
| 7 |
+
---
|
| 8 |
+
# VGen
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
VGen is an open-source video synthesis codebase developed by the Tongyi Lab of Alibaba Group, featuring state-of-the-art video generative models. This repository includes implementations of the following methods:
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
- [I2VGen-xl: High-quality image-to-video synthesis via cascaded diffusion models](https://i2vgen-xl.github.io/)
|
| 17 |
+
- [VideoComposer: Compositional Video Synthesis with Motion Controllability](https://videocomposer.github.io/)
|
| 18 |
+
- [Hierarchical Spatio-temporal Decoupling for Text-to-Video Generation](https://higen-t2v.github.io/)
|
| 19 |
+
- [A Recipe for Scaling up Text-to-Video Generation with Text-free Videos]()
|
| 20 |
+
- [InstructVideo: Instructing Video Diffusion Models with Human Feedback]()
|
| 21 |
+
- [DreamVideo: Composing Your Dream Videos with Customized Subject and Motion](https://dreamvideo-t2v.github.io/)
|
| 22 |
+
- [VideoLCM: Video Latent Consistency Model](https://arxiv.org/abs/2312.09109)
|
| 23 |
+
- [Modelscope text-to-video technical report](https://arxiv.org/abs/2308.06571)
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
VGen can produce high-quality videos from the input text, images, desired motion, desired subjects, and even the feedback signals provided. It also offers a variety of commonly used video generation tools such as visualization, sampling, training, inference, join training using images and videos, acceleration, and more.
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
<a href='https://i2vgen-xl.github.io/'><img src='https://img.shields.io/badge/Project-Page-Green'></a> <a href='https://arxiv.org/abs/2311.04145'><img src='https://img.shields.io/badge/Paper-Arxiv-red'></a> [](https://youtu.be/XUi0y7dxqEQ) <a href='https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441039979087.mp4'><img src='source/logo.png'></a>
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
## 🔥News!!!
|
| 33 |
+
- __[2024.01]__ Diffusers now supports I2VGenXL
|
| 34 |
+
- __[2023.12]__ We release the high-efficiency video generation method [VideoLCM](https://arxiv.org/abs/2312.09109)
|
| 35 |
+
- __[2023.12]__ We release the code and model of I2VGen-XL and the ModelScope T2V
|
| 36 |
+
- __[2023.12]__ We release the T2V method [HiGen](https://higen-t2v.github.io) and customizing T2V method [DreamVideo](https://dreamvideo-t2v.github.io).
|
| 37 |
+
- __[2023.12]__ We write an [introduction docment](doc/introduction.pdf) for VGen and compare I2VGen-XL with SVD.
|
| 38 |
+
- __[2023.11]__ We release a high-quality I2VGen-XL model, please refer to the [Webpage](https://i2vgen-xl.github.io)
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
## TODO
|
| 42 |
+
- [x] Release the technical papers and webpage of [I2VGen-XL](doc/i2vgen-xl.md)
|
| 43 |
+
- [x] Release the code and pretrained models that can generate 1280x720 videos
|
| 44 |
+
- [ ] Release models optimized specifically for the human body and faces
|
| 45 |
+
- [ ] Updated version can fully maintain the ID and capture large and accurate motions simultaneously
|
| 46 |
+
- [ ] Release other methods and the corresponding models
|
| 47 |
+
|
| 48 |
+
|
| 49 |
+
## Preparation
|
| 50 |
+
|
| 51 |
+
The main features of VGen are as follows:
|
| 52 |
+
- Expandability, allowing for easy management of your own experiments.
|
| 53 |
+
- Completeness, encompassing all common components for video generation.
|
| 54 |
+
- Excellent performance, featuring powerful pre-trained models in multiple tasks.
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
### Installation
|
| 58 |
+
|
| 59 |
+
```
|
| 60 |
+
conda create -n vgen python=3.8
|
| 61 |
+
conda activate vgen
|
| 62 |
+
pip install torch==1.12.0+cu113 torchvision==0.13.0+cu113 torchaudio==0.12.0 --extra-index-url https://download.pytorch.org/whl/cu113
|
| 63 |
+
pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
|
| 64 |
+
```
|
| 65 |
+
|
| 66 |
+
### Datasets
|
| 67 |
+
|
| 68 |
+
We have provided a **demo dataset** that includes images and videos, along with their lists in ``data``.
|
| 69 |
+
|
| 70 |
+
*Please note that the demo images used here are for testing purposes and were not included in the training.*
|
| 71 |
+
|
| 72 |
+
|
| 73 |
+
### Clone codeb
|
| 74 |
+
|
| 75 |
+
```
|
| 76 |
+
git clone https://github.com/damo-vilab/i2vgen-xl.git
|
| 77 |
+
cd i2vgen-xl
|
| 78 |
+
```
|
| 79 |
+
|
| 80 |
+
|
| 81 |
+
## Getting Started with VGen
|
| 82 |
+
|
| 83 |
+
### (1) Train your text-to-video model
|
| 84 |
+
|
| 85 |
+
|
| 86 |
+
Executing the following command to enable distributed training is as easy as that.
|
| 87 |
+
```
|
| 88 |
+
python train_net.py --cfg configs/t2v_train.yaml
|
| 89 |
+
```
|
| 90 |
+
|
| 91 |
+
In the `t2v_train.yaml` configuration file, you can specify the data, adjust the video-to-image ratio using `frame_lens`, and validate your ideas with different Diffusion settings, and so on.
|
| 92 |
+
|
| 93 |
+
- Before the training, you can download any of our open-source models for initialization. Our codebase supports custom initialization and `grad_scale` settings, all of which are included in the `Pretrain` item in yaml file.
|
| 94 |
+
- During the training, you can view the saved models and intermediate inference results in the `workspace/experiments/t2v_train`directory.
|
| 95 |
+
|
| 96 |
+
After the training is completed, you can perform inference on the model using the following command.
|
| 97 |
+
```
|
| 98 |
+
python inference.py --cfg configs/t2v_infer.yaml
|
| 99 |
+
```
|
| 100 |
+
Then you can find the videos you generated in the `workspace/experiments/test_img_01` directory. For specific configurations such as data, models, seed, etc., please refer to the `t2v_infer.yaml` file.
|
| 101 |
+
|
| 102 |
+
<!-- <table>
|
| 103 |
+
<center>
|
| 104 |
+
<tr>
|
| 105 |
+
<td ><center>
|
| 106 |
+
<video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441754174077.mp4"></video>
|
| 107 |
+
</center></td>
|
| 108 |
+
<td ><center>
|
| 109 |
+
<video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441138824052.mp4"></video>
|
| 110 |
+
</center></td>
|
| 111 |
+
</tr>
|
| 112 |
+
</center>
|
| 113 |
+
</table>
|
| 114 |
+
</center> -->
|
| 115 |
+
|
| 116 |
+
<table>
|
| 117 |
+
<center>
|
| 118 |
+
<tr>
|
| 119 |
+
<td ><center>
|
| 120 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01Ya2I5I25utrJwJ9Jf_!!6000000007587-2-tps-1280-720.png"></image>
|
| 121 |
+
</center></td>
|
| 122 |
+
<td ><center>
|
| 123 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i3/O1CN01CrmYaz1zXBetmg3dd_!!6000000006723-2-tps-1280-720.png"></image>
|
| 124 |
+
</center></td>
|
| 125 |
+
</tr>
|
| 126 |
+
<tr>
|
| 127 |
+
<td ><center>
|
| 128 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441754174077.mp4">HRER</a> to view the generated video.</p>
|
| 129 |
+
</center></td>
|
| 130 |
+
<td ><center>
|
| 131 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441138824052.mp4">HRER</a> to view the generated video.</p>
|
| 132 |
+
</center></td>
|
| 133 |
+
</tr>
|
| 134 |
+
</center>
|
| 135 |
+
</table>
|
| 136 |
+
</center>
|
| 137 |
+
|
| 138 |
+
|
| 139 |
+
### (2) Run the I2VGen-XL model
|
| 140 |
+
|
| 141 |
+
(i) Download model and test data:
|
| 142 |
+
```
|
| 143 |
+
!pip install modelscope
|
| 144 |
+
from modelscope.hub.snapshot_download import snapshot_download
|
| 145 |
+
model_dir = snapshot_download('damo/I2VGen-XL', cache_dir='models/', revision='v1.0.0')
|
| 146 |
+
```
|
| 147 |
+
|
| 148 |
+
(ii) Run the following command:
|
| 149 |
+
```
|
| 150 |
+
python inference.py --cfg configs/i2vgen_xl_infer.yaml
|
| 151 |
+
```
|
| 152 |
+
In a few minutes, you can retrieve the high-definition video you wish to create from the `workspace/experiments/test_img_01` directory. At present, we find that the current model performs inadequately on **anime images** and **images with a black background** due to the lack of relevant training data. We are consistently working to optimize it.
|
| 153 |
+
|
| 154 |
+
|
| 155 |
+
<span style="color:red">Due to the compression of our video quality in GIF format, please click 'HRER' below to view the original video.</span>
|
| 156 |
+
|
| 157 |
+
<center>
|
| 158 |
+
<table>
|
| 159 |
+
<center>
|
| 160 |
+
<tr>
|
| 161 |
+
<td ><center>
|
| 162 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i1/O1CN01CCEq7K1ZeLpNQqrWu_!!6000000003219-0-tps-1280-720.jpg"></image>
|
| 163 |
+
</center></td>
|
| 164 |
+
<td ><center>
|
| 165 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442125067544.mp4"></video> -->
|
| 166 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01hIQcvG1spmQMLqBo0_!!6000000005816-1-tps-1280-704.gif"></image>
|
| 167 |
+
</center></td>
|
| 168 |
+
</tr>
|
| 169 |
+
<tr>
|
| 170 |
+
<td ><center>
|
| 171 |
+
<p>Input Image</p>
|
| 172 |
+
</center></td>
|
| 173 |
+
<td ><center>
|
| 174 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442125067544.mp4">HRER</a> to view the generated video.</p>
|
| 175 |
+
</center></td>
|
| 176 |
+
</tr>
|
| 177 |
+
<tr>
|
| 178 |
+
<td ><center>
|
| 179 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01ZXY7UN23K8q4oQ3uG_!!6000000007236-2-tps-1280-720.png"></image>
|
| 180 |
+
</center></td>
|
| 181 |
+
<td ><center>
|
| 182 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441385957074.mp4"></video> -->
|
| 183 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i1/O1CN01iaSiiv1aJZURUEY53_!!6000000003309-1-tps-1280-704.gif"></image>
|
| 184 |
+
</center></td>
|
| 185 |
+
</tr>
|
| 186 |
+
<tr>
|
| 187 |
+
<td ><center>
|
| 188 |
+
<p>Input Image</p>
|
| 189 |
+
</center></td>
|
| 190 |
+
<td ><center>
|
| 191 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/441385957074.mp4">HRER</a> to view the generated video.</p>
|
| 192 |
+
</center></td>
|
| 193 |
+
</tr>
|
| 194 |
+
<tr>
|
| 195 |
+
<td ><center>
|
| 196 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i3/O1CN01NHpVGl1oat4H54Hjf_!!6000000005242-2-tps-1280-720.png"></image>
|
| 197 |
+
</center></td>
|
| 198 |
+
<td ><center>
|
| 199 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442102706767.mp4"></video> -->
|
| 200 |
+
<!-- <image muted="true" height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01DgLj1T240jfpzKoaQ_!!6000000007329-1-tps-1280-704.gif"></image>
|
| 201 |
+
-->
|
| 202 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i4/O1CN01DgLj1T240jfpzKoaQ_!!6000000007329-1-tps-1280-704.gif"></image>
|
| 203 |
+
</center></td>
|
| 204 |
+
</tr>
|
| 205 |
+
<tr>
|
| 206 |
+
<td ><center>
|
| 207 |
+
<p>Input Image</p>
|
| 208 |
+
</center></td>
|
| 209 |
+
<td ><center>
|
| 210 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442102706767.mp4">HERE</a> to view the generated video.</p>
|
| 211 |
+
</center></td>
|
| 212 |
+
</tr>
|
| 213 |
+
<tr>
|
| 214 |
+
<td ><center>
|
| 215 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i1/O1CN01odS61s1WW9tXen21S_!!6000000002795-0-tps-1280-720.jpg"></image>
|
| 216 |
+
</center></td>
|
| 217 |
+
<td ><center>
|
| 218 |
+
<!-- <video muted="true" autoplay="true" loop="true" height="260" src="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442163934688.mp4"></video> -->
|
| 219 |
+
<image height="260" src="https://img.alicdn.com/imgextra/i3/O1CN01Jyk1HT28JkZtpAtY6_!!6000000007912-1-tps-1280-704.gif"></image>
|
| 220 |
+
</center></td>
|
| 221 |
+
</tr>
|
| 222 |
+
<tr>
|
| 223 |
+
<td ><center>
|
| 224 |
+
<p>Input Image</p>
|
| 225 |
+
</center></td>
|
| 226 |
+
<td ><center>
|
| 227 |
+
<p>Clike <a href="https://cloud.video.taobao.com/play/u/null/p/1/e/6/t/1/442163934688.mp4">HERE</a> to view the generated video.</p>
|
| 228 |
+
</center></td>
|
| 229 |
+
</tr>
|
| 230 |
+
</center>
|
| 231 |
+
</table>
|
| 232 |
+
</center>
|
| 233 |
+
|
| 234 |
+
### (3) Other methods
|
| 235 |
+
|
| 236 |
+
In preparation.
|
| 237 |
+
|
| 238 |
+
|
| 239 |
+
## Customize your own approach
|
| 240 |
+
|
| 241 |
+
Our codebase essentially supports all the commonly used components in video generation. You can manage your experiments flexibly by adding corresponding registration classes, including `ENGINE, MODEL, DATASETS, EMBEDDER, AUTO_ENCODER, DISTRIBUTION, VISUAL, DIFFUSION, PRETRAIN`, and can be compatible with all our open-source algorithms according to your own needs. If you have any questions, feel free to give us your feedback at any time.
|
| 242 |
+
|
| 243 |
+
## Integration of I2VGenXL with 🧨 diffusers
|
| 244 |
+
|
| 245 |
+
I2VGenXL is supported in the 🧨 diffusers library. Here's how to use it:
|
| 246 |
+
|
| 247 |
+
```python
|
| 248 |
+
import torch
|
| 249 |
+
from diffusers import I2VGenXLPipeline
|
| 250 |
+
from diffusers.utils import load_image, export_to_gif
|
| 251 |
+
|
| 252 |
+
repo_id = "ali-vilab/i2vgen-xl"
|
| 253 |
+
pipeline = I2VGenXLPipeline.from_pretrained(repo_id, torch_dtype=torch.float16, variant="fp16").to("cuda")
|
| 254 |
+
|
| 255 |
+
image_url = "https://github.com/ali-vilab/i2vgen-xl/blob/main/data/test_images/img_0009.png?download=true"
|
| 256 |
+
image = load_image(image_url).convert("RGB")
|
| 257 |
+
prompt = "Papers were floating in the air on a table in the library"
|
| 258 |
+
|
| 259 |
+
generator = torch.manual_seed(8888)
|
| 260 |
+
frames = pipeline(
|
| 261 |
+
prompt=prompt,
|
| 262 |
+
image=image,
|
| 263 |
+
generator=generator
|
| 264 |
+
).frames[0]
|
| 265 |
+
|
| 266 |
+
print(export_to_gif(frames))
|
| 267 |
+
```
|
| 268 |
+
|
| 269 |
+
Find the official documentation [here](https://huggingface.co/docs/diffusers/main/en/api/pipelines/i2vgenxl).
|
| 270 |
+
|
| 271 |
+
Sample output with I2VGenXL:
|
| 272 |
+
|
| 273 |
+
<table>
|
| 274 |
+
<tr>
|
| 275 |
+
<td><center>
|
| 276 |
+
masterpiece, bestquality, sunset.
|
| 277 |
+
<br>
|
| 278 |
+
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/i2vgen-xl-example.gif"
|
| 279 |
+
alt="library"
|
| 280 |
+
style="width: 300px;" />
|
| 281 |
+
</center></td>
|
| 282 |
+
</tr>
|
| 283 |
+
</table>
|
| 284 |
+
|
| 285 |
+
## BibTeX
|
| 286 |
+
|
| 287 |
+
If this repo is useful to you, please cite our corresponding technical paper.
|
| 288 |
+
|
| 289 |
+
|
| 290 |
+
```bibtex
|
| 291 |
+
@article{2023i2vgenxl,
|
| 292 |
+
title={I2VGen-XL: High-Quality Image-to-Video Synthesis via Cascaded Diffusion Models},
|
| 293 |
+
author={Zhang, Shiwei and Wang, Jiayu and Zhang, Yingya and Zhao, Kang and Yuan, Hangjie and Qing, Zhiwu and Wang, Xiang and Zhao, Deli and Zhou, Jingren},
|
| 294 |
+
booktitle={arXiv preprint arXiv:2311.04145},
|
| 295 |
+
year={2023}
|
| 296 |
+
}
|
| 297 |
+
@article{2023videocomposer,
|
| 298 |
+
title={VideoComposer: Compositional Video Synthesis with Motion Controllability},
|
| 299 |
+
author={Wang, Xiang and Yuan, Hangjie and Zhang, Shiwei and Chen, Dayou and Wang, Jiuniu, and Zhang, Yingya, and Shen, Yujun, and Zhao, Deli and Zhou, Jingren},
|
| 300 |
+
booktitle={arXiv preprint arXiv:2306.02018},
|
| 301 |
+
year={2023}
|
| 302 |
+
}
|
| 303 |
+
@article{wang2023modelscope,
|
| 304 |
+
title={Modelscope text-to-video technical report},
|
| 305 |
+
author={Wang, Jiuniu and Yuan, Hangjie and Chen, Dayou and Zhang, Yingya and Wang, Xiang and Zhang, Shiwei},
|
| 306 |
+
journal={arXiv preprint arXiv:2308.06571},
|
| 307 |
+
year={2023}
|
| 308 |
+
}
|
| 309 |
+
@article{dreamvideo,
|
| 310 |
+
title={DreamVideo: Composing Your Dream Videos with Customized Subject and Motion},
|
| 311 |
+
author={Wei, Yujie and Zhang, Shiwei and Qing, Zhiwu and Yuan, Hangjie and Liu, Zhiheng and Liu, Yu and Zhang, Yingya and Zhou, Jingren and Shan, Hongming},
|
| 312 |
+
journal={arXiv preprint arXiv:2312.04433},
|
| 313 |
+
year={2023}
|
| 314 |
+
}
|
| 315 |
+
@article{qing2023higen,
|
| 316 |
+
title={Hierarchical Spatio-temporal Decoupling for Text-to-Video Generation},
|
| 317 |
+
author={Qing, Zhiwu and Zhang, Shiwei and Wang, Jiayu and Wang, Xiang and Wei, Yujie and Zhang, Yingya and Gao, Changxin and Sang, Nong },
|
| 318 |
+
journal={arXiv preprint arXiv:2312.04483},
|
| 319 |
+
year={2023}
|
| 320 |
+
}
|
| 321 |
+
@article{wang2023videolcm,
|
| 322 |
+
title={VideoLCM: Video Latent Consistency Model},
|
| 323 |
+
author={Wang, Xiang and Zhang, Shiwei and Zhang, Han and Liu, Yu and Zhang, Yingya and Gao, Changxin and Sang, Nong },
|
| 324 |
+
journal={arXiv preprint arXiv:2312.09109},
|
| 325 |
+
year={2023}
|
| 326 |
+
}
|
| 327 |
+
```
|
| 328 |
+
|
| 329 |
+
## Disclaimer
|
| 330 |
+
|
| 331 |
+
This open-source model is trained with using [WebVid-10M](https://m-bain.github.io/webvid-dataset/) and [LAION-400M](https://laion.ai/blog/laion-400-open-dataset/) datasets and is intended for <strong>RESEARCH/NON-COMMERCIAL USE ONLY</strong>.
|
| 332 |
+
|
| 333 |
+
|
| 334 |
+
|
doc/.DS_Store
ADDED
|
Binary file (6.15 kB). View file
|
|
|
doc/i2vgen-xl.md
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# I2VGen-XL
|
| 2 |
+
|
| 3 |
+
Official repo for [I2vgen-xl: High-quality image-to-video synthesis via cascaded diffusion models](https://arxiv.org/abs/2311.04145)
|
| 4 |
+
|
| 5 |
+
Please see [Project Page](https://i2vgen-xl.github.io) for more examples.
|
| 6 |
+
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
I2VGen-XL is capable of generating high-quality, realistically animated, and temporally coherent high-definition videos from a single input static image, based on user input.
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
*Our initial version has already been open-sourced on [Modelscope](https://modelscope.cn/models/damo/Image-to-Video/summary). This project focuses on improving the version, especially in terms of motions and semantics.*
|
| 15 |
+
|
| 16 |
+
## Examples
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
|
doc/introduction.pdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f4d416283eb95212e1fd45c2d02045a836160929fd15e7120dd77998380c7656
|
| 3 |
+
size 4857845
|
feature_extractor/preprocessor_config.json
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"crop_size": {
|
| 3 |
+
"height": 224,
|
| 4 |
+
"width": 224
|
| 5 |
+
},
|
| 6 |
+
"do_center_crop": true,
|
| 7 |
+
"do_convert_rgb": true,
|
| 8 |
+
"do_normalize": true,
|
| 9 |
+
"do_rescale": true,
|
| 10 |
+
"do_resize": true,
|
| 11 |
+
"image_mean": [
|
| 12 |
+
0.48145466,
|
| 13 |
+
0.4578275,
|
| 14 |
+
0.40821073
|
| 15 |
+
],
|
| 16 |
+
"image_processor_type": "CLIPImageProcessor",
|
| 17 |
+
"image_std": [
|
| 18 |
+
0.26862954,
|
| 19 |
+
0.26130258,
|
| 20 |
+
0.27577711
|
| 21 |
+
],
|
| 22 |
+
"resample": 3,
|
| 23 |
+
"rescale_factor": 0.00392156862745098,
|
| 24 |
+
"size": {
|
| 25 |
+
"shortest_edge": 224
|
| 26 |
+
}
|
| 27 |
+
}
|
image_encoder/config.json
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "i2vgen-xl/image_encoder",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"CLIPVisionModelWithProjection"
|
| 5 |
+
],
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"dropout": 0.0,
|
| 8 |
+
"hidden_act": "gelu",
|
| 9 |
+
"hidden_size": 1280,
|
| 10 |
+
"image_size": 224,
|
| 11 |
+
"initializer_factor": 1.0,
|
| 12 |
+
"initializer_range": 0.02,
|
| 13 |
+
"intermediate_size": 5120,
|
| 14 |
+
"layer_norm_eps": 1e-05,
|
| 15 |
+
"model_type": "clip_vision_model",
|
| 16 |
+
"num_attention_heads": 16,
|
| 17 |
+
"num_channels": 3,
|
| 18 |
+
"num_hidden_layers": 32,
|
| 19 |
+
"patch_size": 14,
|
| 20 |
+
"projection_dim": 1024,
|
| 21 |
+
"torch_dtype": "float16",
|
| 22 |
+
"transformers_version": "4.36.2"
|
| 23 |
+
}
|
image_encoder/model.fp16.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ae616c24393dd1854372b0639e5541666f7521cbe219669255e865cb7f89466a
|
| 3 |
+
size 1264217240
|
image_encoder/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ed1e5af7b4042ca30ec29999a4a5cfcac90b7fb610fd05ace834f2dcbb763eab
|
| 3 |
+
size 2528371296
|
model_index.json
ADDED
|
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_class_name": "I2VGenXLPipeline",
|
| 3 |
+
"_diffusers_version": "0.26.1",
|
| 4 |
+
"_name_or_path": "i2vgen-xl",
|
| 5 |
+
"feature_extractor": [
|
| 6 |
+
"transformers",
|
| 7 |
+
"CLIPImageProcessor"
|
| 8 |
+
],
|
| 9 |
+
"image_encoder": [
|
| 10 |
+
"transformers",
|
| 11 |
+
"CLIPVisionModelWithProjection"
|
| 12 |
+
],
|
| 13 |
+
"scheduler": [
|
| 14 |
+
"diffusers",
|
| 15 |
+
"DDIMScheduler"
|
| 16 |
+
],
|
| 17 |
+
"text_encoder": [
|
| 18 |
+
"transformers",
|
| 19 |
+
"CLIPTextModel"
|
| 20 |
+
],
|
| 21 |
+
"tokenizer": [
|
| 22 |
+
"transformers",
|
| 23 |
+
"CLIPTokenizer"
|
| 24 |
+
],
|
| 25 |
+
"unet": [
|
| 26 |
+
"diffusers",
|
| 27 |
+
"I2VGenXLUNet"
|
| 28 |
+
],
|
| 29 |
+
"vae": [
|
| 30 |
+
"diffusers",
|
| 31 |
+
"AutoencoderKL"
|
| 32 |
+
]
|
| 33 |
+
}
|
models/i2vgen_xl_00854500.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d3921efea21a4aac109a03ea8ca1f4f6a756ba82d30bfeb82b83f94a7aff8f73
|
| 3 |
+
size 5682502260
|
models/open_clip_pytorch_model.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9a78ef8e8c73fd0df621682e7a8e8eb36c6916cb3c16b291a082ecd52ab79cc4
|
| 3 |
+
size 3944692325
|
models/stable_diffusion_image_key_temporal_attention_x1.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
["input_blocks.0.0.weight", "input_blocks.0.0.bias", "input_blocks.1.0.in_layers.0.weight", "input_blocks.1.0.in_layers.0.bias", "input_blocks.1.0.in_layers.2.weight", "input_blocks.1.0.in_layers.2.bias", "input_blocks.1.0.emb_layers.1.weight", "input_blocks.1.0.emb_layers.1.bias", "input_blocks.1.0.out_layers.0.weight", "input_blocks.1.0.out_layers.0.bias", "input_blocks.1.0.out_layers.3.weight", "input_blocks.1.0.out_layers.3.bias", "input_blocks.1.1.norm.weight", "input_blocks.1.1.norm.bias", "input_blocks.1.1.proj_in.weight", "input_blocks.1.1.proj_in.bias", "input_blocks.1.1.transformer_blocks.0.attn1.to_q.weight", "input_blocks.1.1.transformer_blocks.0.attn1.to_k.weight", "input_blocks.1.1.transformer_blocks.0.attn1.to_v.weight", "input_blocks.1.1.transformer_blocks.0.attn1.to_out.0.weight", "input_blocks.1.1.transformer_blocks.0.attn1.to_out.0.bias", "input_blocks.1.1.transformer_blocks.0.ff.net.0.proj.weight", "input_blocks.1.1.transformer_blocks.0.ff.net.0.proj.bias", "input_blocks.1.1.transformer_blocks.0.ff.net.2.weight", "input_blocks.1.1.transformer_blocks.0.ff.net.2.bias", "input_blocks.1.1.transformer_blocks.0.attn2.to_q.weight", "input_blocks.1.1.transformer_blocks.0.attn2.to_k.weight", "input_blocks.1.1.transformer_blocks.0.attn2.to_v.weight", "input_blocks.1.1.transformer_blocks.0.attn2.to_out.0.weight", "input_blocks.1.1.transformer_blocks.0.attn2.to_out.0.bias", "input_blocks.1.1.transformer_blocks.0.norm1.weight", "input_blocks.1.1.transformer_blocks.0.norm1.bias", "input_blocks.1.1.transformer_blocks.0.norm2.weight", "input_blocks.1.1.transformer_blocks.0.norm2.bias", "input_blocks.1.1.transformer_blocks.0.norm3.weight", "input_blocks.1.1.transformer_blocks.0.norm3.bias", "input_blocks.1.1.proj_out.weight", "input_blocks.1.1.proj_out.bias", "input_blocks.2.0.in_layers.0.weight", "input_blocks.2.0.in_layers.0.bias", "input_blocks.2.0.in_layers.2.weight", "input_blocks.2.0.in_layers.2.bias", "input_blocks.2.0.emb_layers.1.weight", "input_blocks.2.0.emb_layers.1.bias", "input_blocks.2.0.out_layers.0.weight", "input_blocks.2.0.out_layers.0.bias", "input_blocks.2.0.out_layers.3.weight", "input_blocks.2.0.out_layers.3.bias", "input_blocks.2.1.norm.weight", "input_blocks.2.1.norm.bias", "input_blocks.2.1.proj_in.weight", "input_blocks.2.1.proj_in.bias", "input_blocks.2.1.transformer_blocks.0.attn1.to_q.weight", "input_blocks.2.1.transformer_blocks.0.attn1.to_k.weight", "input_blocks.2.1.transformer_blocks.0.attn1.to_v.weight", "input_blocks.2.1.transformer_blocks.0.attn1.to_out.0.weight", "input_blocks.2.1.transformer_blocks.0.attn1.to_out.0.bias", "input_blocks.2.1.transformer_blocks.0.ff.net.0.proj.weight", "input_blocks.2.1.transformer_blocks.0.ff.net.0.proj.bias", "input_blocks.2.1.transformer_blocks.0.ff.net.2.weight", "input_blocks.2.1.transformer_blocks.0.ff.net.2.bias", "input_blocks.2.1.transformer_blocks.0.attn2.to_q.weight", "input_blocks.2.1.transformer_blocks.0.attn2.to_k.weight", "input_blocks.2.1.transformer_blocks.0.attn2.to_v.weight", "input_blocks.2.1.transformer_blocks.0.attn2.to_out.0.weight", "input_blocks.2.1.transformer_blocks.0.attn2.to_out.0.bias", "input_blocks.2.1.transformer_blocks.0.norm1.weight", "input_blocks.2.1.transformer_blocks.0.norm1.bias", "input_blocks.2.1.transformer_blocks.0.norm2.weight", "input_blocks.2.1.transformer_blocks.0.norm2.bias", "input_blocks.2.1.transformer_blocks.0.norm3.weight", "input_blocks.2.1.transformer_blocks.0.norm3.bias", "input_blocks.2.1.proj_out.weight", "input_blocks.2.1.proj_out.bias", "input_blocks.3.op.weight", "input_blocks.3.op.bias", "input_blocks.4.0.in_layers.0.weight", "input_blocks.4.0.in_layers.0.bias", "input_blocks.4.0.in_layers.2.weight", "input_blocks.4.0.in_layers.2.bias", "input_blocks.4.0.emb_layers.1.weight", "input_blocks.4.0.emb_layers.1.bias", "input_blocks.4.0.out_layers.0.weight", "input_blocks.4.0.out_layers.0.bias", "input_blocks.4.0.out_layers.3.weight", "input_blocks.4.0.out_layers.3.bias", "input_blocks.4.0.skip_connection.weight", "input_blocks.4.0.skip_connection.bias", "input_blocks.4.1.norm.weight", "input_blocks.4.1.norm.bias", "input_blocks.4.1.proj_in.weight", "input_blocks.4.1.proj_in.bias", "input_blocks.4.1.transformer_blocks.0.attn1.to_q.weight", "input_blocks.4.1.transformer_blocks.0.attn1.to_k.weight", "input_blocks.4.1.transformer_blocks.0.attn1.to_v.weight", "input_blocks.4.1.transformer_blocks.0.attn1.to_out.0.weight", "input_blocks.4.1.transformer_blocks.0.attn1.to_out.0.bias", "input_blocks.4.1.transformer_blocks.0.ff.net.0.proj.weight", "input_blocks.4.1.transformer_blocks.0.ff.net.0.proj.bias", "input_blocks.4.1.transformer_blocks.0.ff.net.2.weight", "input_blocks.4.1.transformer_blocks.0.ff.net.2.bias", "input_blocks.4.1.transformer_blocks.0.attn2.to_q.weight", "input_blocks.4.1.transformer_blocks.0.attn2.to_k.weight", "input_blocks.4.1.transformer_blocks.0.attn2.to_v.weight", "input_blocks.4.1.transformer_blocks.0.attn2.to_out.0.weight", "input_blocks.4.1.transformer_blocks.0.attn2.to_out.0.bias", "input_blocks.4.1.transformer_blocks.0.norm1.weight", "input_blocks.4.1.transformer_blocks.0.norm1.bias", "input_blocks.4.1.transformer_blocks.0.norm2.weight", "input_blocks.4.1.transformer_blocks.0.norm2.bias", "input_blocks.4.1.transformer_blocks.0.norm3.weight", "input_blocks.4.1.transformer_blocks.0.norm3.bias", "input_blocks.4.1.proj_out.weight", "input_blocks.4.1.proj_out.bias", "input_blocks.5.0.in_layers.0.weight", "input_blocks.5.0.in_layers.0.bias", "input_blocks.5.0.in_layers.2.weight", "input_blocks.5.0.in_layers.2.bias", "input_blocks.5.0.emb_layers.1.weight", "input_blocks.5.0.emb_layers.1.bias", "input_blocks.5.0.out_layers.0.weight", "input_blocks.5.0.out_layers.0.bias", "input_blocks.5.0.out_layers.3.weight", "input_blocks.5.0.out_layers.3.bias", "input_blocks.5.1.norm.weight", "input_blocks.5.1.norm.bias", "input_blocks.5.1.proj_in.weight", "input_blocks.5.1.proj_in.bias", "input_blocks.5.1.transformer_blocks.0.attn1.to_q.weight", "input_blocks.5.1.transformer_blocks.0.attn1.to_k.weight", "input_blocks.5.1.transformer_blocks.0.attn1.to_v.weight", "input_blocks.5.1.transformer_blocks.0.attn1.to_out.0.weight", "input_blocks.5.1.transformer_blocks.0.attn1.to_out.0.bias", "input_blocks.5.1.transformer_blocks.0.ff.net.0.proj.weight", "input_blocks.5.1.transformer_blocks.0.ff.net.0.proj.bias", "input_blocks.5.1.transformer_blocks.0.ff.net.2.weight", "input_blocks.5.1.transformer_blocks.0.ff.net.2.bias", "input_blocks.5.1.transformer_blocks.0.attn2.to_q.weight", "input_blocks.5.1.transformer_blocks.0.attn2.to_k.weight", "input_blocks.5.1.transformer_blocks.0.attn2.to_v.weight", "input_blocks.5.1.transformer_blocks.0.attn2.to_out.0.weight", "input_blocks.5.1.transformer_blocks.0.attn2.to_out.0.bias", "input_blocks.5.1.transformer_blocks.0.norm1.weight", "input_blocks.5.1.transformer_blocks.0.norm1.bias", "input_blocks.5.1.transformer_blocks.0.norm2.weight", "input_blocks.5.1.transformer_blocks.0.norm2.bias", "input_blocks.5.1.transformer_blocks.0.norm3.weight", "input_blocks.5.1.transformer_blocks.0.norm3.bias", "input_blocks.5.1.proj_out.weight", "input_blocks.5.1.proj_out.bias", "input_blocks.6.op.weight", "input_blocks.6.op.bias", "input_blocks.7.0.in_layers.0.weight", "input_blocks.7.0.in_layers.0.bias", "input_blocks.7.0.in_layers.2.weight", "input_blocks.7.0.in_layers.2.bias", "input_blocks.7.0.emb_layers.1.weight", "input_blocks.7.0.emb_layers.1.bias", "input_blocks.7.0.out_layers.0.weight", "input_blocks.7.0.out_layers.0.bias", "input_blocks.7.0.out_layers.3.weight", "input_blocks.7.0.out_layers.3.bias", "input_blocks.7.0.skip_connection.weight", "input_blocks.7.0.skip_connection.bias", "input_blocks.7.1.norm.weight", "input_blocks.7.1.norm.bias", "input_blocks.7.1.proj_in.weight", "input_blocks.7.1.proj_in.bias", "input_blocks.7.1.transformer_blocks.0.attn1.to_q.weight", "input_blocks.7.1.transformer_blocks.0.attn1.to_k.weight", "input_blocks.7.1.transformer_blocks.0.attn1.to_v.weight", "input_blocks.7.1.transformer_blocks.0.attn1.to_out.0.weight", "input_blocks.7.1.transformer_blocks.0.attn1.to_out.0.bias", "input_blocks.7.1.transformer_blocks.0.ff.net.0.proj.weight", "input_blocks.7.1.transformer_blocks.0.ff.net.0.proj.bias", "input_blocks.7.1.transformer_blocks.0.ff.net.2.weight", "input_blocks.7.1.transformer_blocks.0.ff.net.2.bias", "input_blocks.7.1.transformer_blocks.0.attn2.to_q.weight", "input_blocks.7.1.transformer_blocks.0.attn2.to_k.weight", "input_blocks.7.1.transformer_blocks.0.attn2.to_v.weight", "input_blocks.7.1.transformer_blocks.0.attn2.to_out.0.weight", "input_blocks.7.1.transformer_blocks.0.attn2.to_out.0.bias", "input_blocks.7.1.transformer_blocks.0.norm1.weight", "input_blocks.7.1.transformer_blocks.0.norm1.bias", "input_blocks.7.1.transformer_blocks.0.norm2.weight", "input_blocks.7.1.transformer_blocks.0.norm2.bias", "input_blocks.7.1.transformer_blocks.0.norm3.weight", "input_blocks.7.1.transformer_blocks.0.norm3.bias", "input_blocks.7.1.proj_out.weight", "input_blocks.7.1.proj_out.bias", "input_blocks.8.0.in_layers.0.weight", "input_blocks.8.0.in_layers.0.bias", "input_blocks.8.0.in_layers.2.weight", "input_blocks.8.0.in_layers.2.bias", "input_blocks.8.0.emb_layers.1.weight", "input_blocks.8.0.emb_layers.1.bias", "input_blocks.8.0.out_layers.0.weight", "input_blocks.8.0.out_layers.0.bias", "input_blocks.8.0.out_layers.3.weight", "input_blocks.8.0.out_layers.3.bias", "input_blocks.8.1.norm.weight", "input_blocks.8.1.norm.bias", "input_blocks.8.1.proj_in.weight", "input_blocks.8.1.proj_in.bias", "input_blocks.8.1.transformer_blocks.0.attn1.to_q.weight", "input_blocks.8.1.transformer_blocks.0.attn1.to_k.weight", "input_blocks.8.1.transformer_blocks.0.attn1.to_v.weight", "input_blocks.8.1.transformer_blocks.0.attn1.to_out.0.weight", "input_blocks.8.1.transformer_blocks.0.attn1.to_out.0.bias", "input_blocks.8.1.transformer_blocks.0.ff.net.0.proj.weight", "input_blocks.8.1.transformer_blocks.0.ff.net.0.proj.bias", "input_blocks.8.1.transformer_blocks.0.ff.net.2.weight", "input_blocks.8.1.transformer_blocks.0.ff.net.2.bias", "input_blocks.8.1.transformer_blocks.0.attn2.to_q.weight", "input_blocks.8.1.transformer_blocks.0.attn2.to_k.weight", "input_blocks.8.1.transformer_blocks.0.attn2.to_v.weight", "input_blocks.8.1.transformer_blocks.0.attn2.to_out.0.weight", "input_blocks.8.1.transformer_blocks.0.attn2.to_out.0.bias", "input_blocks.8.1.transformer_blocks.0.norm1.weight", "input_blocks.8.1.transformer_blocks.0.norm1.bias", "input_blocks.8.1.transformer_blocks.0.norm2.weight", "input_blocks.8.1.transformer_blocks.0.norm2.bias", "input_blocks.8.1.transformer_blocks.0.norm3.weight", "input_blocks.8.1.transformer_blocks.0.norm3.bias", "input_blocks.8.1.proj_out.weight", "input_blocks.8.1.proj_out.bias", "input_blocks.9.op.weight", "input_blocks.9.op.bias", "input_blocks.10.0.in_layers.0.weight", "input_blocks.10.0.in_layers.0.bias", "input_blocks.10.0.in_layers.2.weight", "input_blocks.10.0.in_layers.2.bias", "input_blocks.10.0.emb_layers.1.weight", "input_blocks.10.0.emb_layers.1.bias", "input_blocks.10.0.out_layers.0.weight", "input_blocks.10.0.out_layers.0.bias", "input_blocks.10.0.out_layers.3.weight", "input_blocks.10.0.out_layers.3.bias", "input_blocks.11.0.in_layers.0.weight", "input_blocks.11.0.in_layers.0.bias", "input_blocks.11.0.in_layers.2.weight", "input_blocks.11.0.in_layers.2.bias", "input_blocks.11.0.emb_layers.1.weight", "input_blocks.11.0.emb_layers.1.bias", "input_blocks.11.0.out_layers.0.weight", "input_blocks.11.0.out_layers.0.bias", "input_blocks.11.0.out_layers.3.weight", "input_blocks.11.0.out_layers.3.bias", "middle_block.0.in_layers.0.weight", "middle_block.0.in_layers.0.bias", "middle_block.0.in_layers.2.weight", "middle_block.0.in_layers.2.bias", "middle_block.0.emb_layers.1.weight", "middle_block.0.emb_layers.1.bias", "middle_block.0.out_layers.0.weight", "middle_block.0.out_layers.0.bias", "middle_block.0.out_layers.3.weight", "middle_block.0.out_layers.3.bias", "middle_block.1.norm.weight", "middle_block.1.norm.bias", "middle_block.1.proj_in.weight", "middle_block.1.proj_in.bias", "middle_block.1.transformer_blocks.0.attn1.to_q.weight", "middle_block.1.transformer_blocks.0.attn1.to_k.weight", "middle_block.1.transformer_blocks.0.attn1.to_v.weight", "middle_block.1.transformer_blocks.0.attn1.to_out.0.weight", "middle_block.1.transformer_blocks.0.attn1.to_out.0.bias", "middle_block.1.transformer_blocks.0.ff.net.0.proj.weight", "middle_block.1.transformer_blocks.0.ff.net.0.proj.bias", "middle_block.1.transformer_blocks.0.ff.net.2.weight", "middle_block.1.transformer_blocks.0.ff.net.2.bias", "middle_block.1.transformer_blocks.0.attn2.to_q.weight", "middle_block.1.transformer_blocks.0.attn2.to_k.weight", "middle_block.1.transformer_blocks.0.attn2.to_v.weight", "middle_block.1.transformer_blocks.0.attn2.to_out.0.weight", "middle_block.1.transformer_blocks.0.attn2.to_out.0.bias", "middle_block.1.transformer_blocks.0.norm1.weight", "middle_block.1.transformer_blocks.0.norm1.bias", "middle_block.1.transformer_blocks.0.norm2.weight", "middle_block.1.transformer_blocks.0.norm2.bias", "middle_block.1.transformer_blocks.0.norm3.weight", "middle_block.1.transformer_blocks.0.norm3.bias", "middle_block.1.proj_out.weight", "middle_block.1.proj_out.bias", "middle_block.3.in_layers.0.weight", "middle_block.3.in_layers.0.bias", "middle_block.3.in_layers.2.weight", "middle_block.3.in_layers.2.bias", "middle_block.3.emb_layers.1.weight", "middle_block.3.emb_layers.1.bias", "middle_block.3.out_layers.0.weight", "middle_block.3.out_layers.0.bias", "middle_block.3.out_layers.3.weight", "middle_block.3.out_layers.3.bias", "output_blocks.0.0.in_layers.0.weight", "output_blocks.0.0.in_layers.0.bias", "output_blocks.0.0.in_layers.2.weight", "output_blocks.0.0.in_layers.2.bias", "output_blocks.0.0.emb_layers.1.weight", "output_blocks.0.0.emb_layers.1.bias", "output_blocks.0.0.out_layers.0.weight", "output_blocks.0.0.out_layers.0.bias", "output_blocks.0.0.out_layers.3.weight", "output_blocks.0.0.out_layers.3.bias", "output_blocks.0.0.skip_connection.weight", "output_blocks.0.0.skip_connection.bias", "output_blocks.1.0.in_layers.0.weight", "output_blocks.1.0.in_layers.0.bias", "output_blocks.1.0.in_layers.2.weight", "output_blocks.1.0.in_layers.2.bias", "output_blocks.1.0.emb_layers.1.weight", "output_blocks.1.0.emb_layers.1.bias", "output_blocks.1.0.out_layers.0.weight", "output_blocks.1.0.out_layers.0.bias", "output_blocks.1.0.out_layers.3.weight", "output_blocks.1.0.out_layers.3.bias", "output_blocks.1.0.skip_connection.weight", "output_blocks.1.0.skip_connection.bias", "output_blocks.2.0.in_layers.0.weight", "output_blocks.2.0.in_layers.0.bias", "output_blocks.2.0.in_layers.2.weight", "output_blocks.2.0.in_layers.2.bias", "output_blocks.2.0.emb_layers.1.weight", "output_blocks.2.0.emb_layers.1.bias", "output_blocks.2.0.out_layers.0.weight", "output_blocks.2.0.out_layers.0.bias", "output_blocks.2.0.out_layers.3.weight", "output_blocks.2.0.out_layers.3.bias", "output_blocks.2.0.skip_connection.weight", "output_blocks.2.0.skip_connection.bias", "output_blocks.2.1.conv.weight", "output_blocks.2.1.conv.bias", "output_blocks.3.0.in_layers.0.weight", "output_blocks.3.0.in_layers.0.bias", "output_blocks.3.0.in_layers.2.weight", "output_blocks.3.0.in_layers.2.bias", "output_blocks.3.0.emb_layers.1.weight", "output_blocks.3.0.emb_layers.1.bias", "output_blocks.3.0.out_layers.0.weight", "output_blocks.3.0.out_layers.0.bias", "output_blocks.3.0.out_layers.3.weight", "output_blocks.3.0.out_layers.3.bias", "output_blocks.3.0.skip_connection.weight", "output_blocks.3.0.skip_connection.bias", "output_blocks.3.1.norm.weight", "output_blocks.3.1.norm.bias", "output_blocks.3.1.proj_in.weight", "output_blocks.3.1.proj_in.bias", "output_blocks.3.1.transformer_blocks.0.attn1.to_q.weight", "output_blocks.3.1.transformer_blocks.0.attn1.to_k.weight", "output_blocks.3.1.transformer_blocks.0.attn1.to_v.weight", "output_blocks.3.1.transformer_blocks.0.attn1.to_out.0.weight", "output_blocks.3.1.transformer_blocks.0.attn1.to_out.0.bias", "output_blocks.3.1.transformer_blocks.0.ff.net.0.proj.weight", "output_blocks.3.1.transformer_blocks.0.ff.net.0.proj.bias", "output_blocks.3.1.transformer_blocks.0.ff.net.2.weight", "output_blocks.3.1.transformer_blocks.0.ff.net.2.bias", "output_blocks.3.1.transformer_blocks.0.attn2.to_q.weight", "output_blocks.3.1.transformer_blocks.0.attn2.to_k.weight", "output_blocks.3.1.transformer_blocks.0.attn2.to_v.weight", "output_blocks.3.1.transformer_blocks.0.attn2.to_out.0.weight", "output_blocks.3.1.transformer_blocks.0.attn2.to_out.0.bias", "output_blocks.3.1.transformer_blocks.0.norm1.weight", "output_blocks.3.1.transformer_blocks.0.norm1.bias", "output_blocks.3.1.transformer_blocks.0.norm2.weight", "output_blocks.3.1.transformer_blocks.0.norm2.bias", "output_blocks.3.1.transformer_blocks.0.norm3.weight", "output_blocks.3.1.transformer_blocks.0.norm3.bias", "output_blocks.3.1.proj_out.weight", "output_blocks.3.1.proj_out.bias", "output_blocks.4.0.in_layers.0.weight", "output_blocks.4.0.in_layers.0.bias", "output_blocks.4.0.in_layers.2.weight", "output_blocks.4.0.in_layers.2.bias", "output_blocks.4.0.emb_layers.1.weight", "output_blocks.4.0.emb_layers.1.bias", "output_blocks.4.0.out_layers.0.weight", "output_blocks.4.0.out_layers.0.bias", "output_blocks.4.0.out_layers.3.weight", "output_blocks.4.0.out_layers.3.bias", "output_blocks.4.0.skip_connection.weight", "output_blocks.4.0.skip_connection.bias", "output_blocks.4.1.norm.weight", "output_blocks.4.1.norm.bias", "output_blocks.4.1.proj_in.weight", "output_blocks.4.1.proj_in.bias", "output_blocks.4.1.transformer_blocks.0.attn1.to_q.weight", "output_blocks.4.1.transformer_blocks.0.attn1.to_k.weight", "output_blocks.4.1.transformer_blocks.0.attn1.to_v.weight", "output_blocks.4.1.transformer_blocks.0.attn1.to_out.0.weight", "output_blocks.4.1.transformer_blocks.0.attn1.to_out.0.bias", "output_blocks.4.1.transformer_blocks.0.ff.net.0.proj.weight", "output_blocks.4.1.transformer_blocks.0.ff.net.0.proj.bias", "output_blocks.4.1.transformer_blocks.0.ff.net.2.weight", "output_blocks.4.1.transformer_blocks.0.ff.net.2.bias", "output_blocks.4.1.transformer_blocks.0.attn2.to_q.weight", "output_blocks.4.1.transformer_blocks.0.attn2.to_k.weight", "output_blocks.4.1.transformer_blocks.0.attn2.to_v.weight", "output_blocks.4.1.transformer_blocks.0.attn2.to_out.0.weight", "output_blocks.4.1.transformer_blocks.0.attn2.to_out.0.bias", "output_blocks.4.1.transformer_blocks.0.norm1.weight", "output_blocks.4.1.transformer_blocks.0.norm1.bias", "output_blocks.4.1.transformer_blocks.0.norm2.weight", "output_blocks.4.1.transformer_blocks.0.norm2.bias", "output_blocks.4.1.transformer_blocks.0.norm3.weight", "output_blocks.4.1.transformer_blocks.0.norm3.bias", "output_blocks.4.1.proj_out.weight", "output_blocks.4.1.proj_out.bias", "output_blocks.5.0.in_layers.0.weight", "output_blocks.5.0.in_layers.0.bias", "output_blocks.5.0.in_layers.2.weight", "output_blocks.5.0.in_layers.2.bias", "output_blocks.5.0.emb_layers.1.weight", "output_blocks.5.0.emb_layers.1.bias", "output_blocks.5.0.out_layers.0.weight", "output_blocks.5.0.out_layers.0.bias", "output_blocks.5.0.out_layers.3.weight", "output_blocks.5.0.out_layers.3.bias", "output_blocks.5.0.skip_connection.weight", "output_blocks.5.0.skip_connection.bias", "output_blocks.5.1.norm.weight", "output_blocks.5.1.norm.bias", "output_blocks.5.1.proj_in.weight", "output_blocks.5.1.proj_in.bias", "output_blocks.5.1.transformer_blocks.0.attn1.to_q.weight", "output_blocks.5.1.transformer_blocks.0.attn1.to_k.weight", "output_blocks.5.1.transformer_blocks.0.attn1.to_v.weight", "output_blocks.5.1.transformer_blocks.0.attn1.to_out.0.weight", "output_blocks.5.1.transformer_blocks.0.attn1.to_out.0.bias", "output_blocks.5.1.transformer_blocks.0.ff.net.0.proj.weight", "output_blocks.5.1.transformer_blocks.0.ff.net.0.proj.bias", "output_blocks.5.1.transformer_blocks.0.ff.net.2.weight", "output_blocks.5.1.transformer_blocks.0.ff.net.2.bias", "output_blocks.5.1.transformer_blocks.0.attn2.to_q.weight", "output_blocks.5.1.transformer_blocks.0.attn2.to_k.weight", "output_blocks.5.1.transformer_blocks.0.attn2.to_v.weight", "output_blocks.5.1.transformer_blocks.0.attn2.to_out.0.weight", "output_blocks.5.1.transformer_blocks.0.attn2.to_out.0.bias", "output_blocks.5.1.transformer_blocks.0.norm1.weight", "output_blocks.5.1.transformer_blocks.0.norm1.bias", "output_blocks.5.1.transformer_blocks.0.norm2.weight", "output_blocks.5.1.transformer_blocks.0.norm2.bias", "output_blocks.5.1.transformer_blocks.0.norm3.weight", "output_blocks.5.1.transformer_blocks.0.norm3.bias", "output_blocks.5.1.proj_out.weight", "output_blocks.5.1.proj_out.bias", "output_blocks.5.3.conv.weight", "output_blocks.5.3.conv.bias", "output_blocks.6.0.in_layers.0.weight", "output_blocks.6.0.in_layers.0.bias", "output_blocks.6.0.in_layers.2.weight", "output_blocks.6.0.in_layers.2.bias", "output_blocks.6.0.emb_layers.1.weight", "output_blocks.6.0.emb_layers.1.bias", "output_blocks.6.0.out_layers.0.weight", "output_blocks.6.0.out_layers.0.bias", "output_blocks.6.0.out_layers.3.weight", "output_blocks.6.0.out_layers.3.bias", "output_blocks.6.0.skip_connection.weight", "output_blocks.6.0.skip_connection.bias", "output_blocks.6.1.norm.weight", "output_blocks.6.1.norm.bias", "output_blocks.6.1.proj_in.weight", "output_blocks.6.1.proj_in.bias", "output_blocks.6.1.transformer_blocks.0.attn1.to_q.weight", "output_blocks.6.1.transformer_blocks.0.attn1.to_k.weight", "output_blocks.6.1.transformer_blocks.0.attn1.to_v.weight", "output_blocks.6.1.transformer_blocks.0.attn1.to_out.0.weight", "output_blocks.6.1.transformer_blocks.0.attn1.to_out.0.bias", "output_blocks.6.1.transformer_blocks.0.ff.net.0.proj.weight", "output_blocks.6.1.transformer_blocks.0.ff.net.0.proj.bias", "output_blocks.6.1.transformer_blocks.0.ff.net.2.weight", "output_blocks.6.1.transformer_blocks.0.ff.net.2.bias", "output_blocks.6.1.transformer_blocks.0.attn2.to_q.weight", "output_blocks.6.1.transformer_blocks.0.attn2.to_k.weight", "output_blocks.6.1.transformer_blocks.0.attn2.to_v.weight", "output_blocks.6.1.transformer_blocks.0.attn2.to_out.0.weight", "output_blocks.6.1.transformer_blocks.0.attn2.to_out.0.bias", "output_blocks.6.1.transformer_blocks.0.norm1.weight", "output_blocks.6.1.transformer_blocks.0.norm1.bias", "output_blocks.6.1.transformer_blocks.0.norm2.weight", "output_blocks.6.1.transformer_blocks.0.norm2.bias", "output_blocks.6.1.transformer_blocks.0.norm3.weight", "output_blocks.6.1.transformer_blocks.0.norm3.bias", "output_blocks.6.1.proj_out.weight", "output_blocks.6.1.proj_out.bias", "output_blocks.7.0.in_layers.0.weight", "output_blocks.7.0.in_layers.0.bias", "output_blocks.7.0.in_layers.2.weight", "output_blocks.7.0.in_layers.2.bias", "output_blocks.7.0.emb_layers.1.weight", "output_blocks.7.0.emb_layers.1.bias", "output_blocks.7.0.out_layers.0.weight", "output_blocks.7.0.out_layers.0.bias", "output_blocks.7.0.out_layers.3.weight", "output_blocks.7.0.out_layers.3.bias", "output_blocks.7.0.skip_connection.weight", "output_blocks.7.0.skip_connection.bias", "output_blocks.7.1.norm.weight", "output_blocks.7.1.norm.bias", "output_blocks.7.1.proj_in.weight", "output_blocks.7.1.proj_in.bias", "output_blocks.7.1.transformer_blocks.0.attn1.to_q.weight", "output_blocks.7.1.transformer_blocks.0.attn1.to_k.weight", "output_blocks.7.1.transformer_blocks.0.attn1.to_v.weight", "output_blocks.7.1.transformer_blocks.0.attn1.to_out.0.weight", "output_blocks.7.1.transformer_blocks.0.attn1.to_out.0.bias", "output_blocks.7.1.transformer_blocks.0.ff.net.0.proj.weight", "output_blocks.7.1.transformer_blocks.0.ff.net.0.proj.bias", "output_blocks.7.1.transformer_blocks.0.ff.net.2.weight", "output_blocks.7.1.transformer_blocks.0.ff.net.2.bias", "output_blocks.7.1.transformer_blocks.0.attn2.to_q.weight", "output_blocks.7.1.transformer_blocks.0.attn2.to_k.weight", "output_blocks.7.1.transformer_blocks.0.attn2.to_v.weight", "output_blocks.7.1.transformer_blocks.0.attn2.to_out.0.weight", "output_blocks.7.1.transformer_blocks.0.attn2.to_out.0.bias", "output_blocks.7.1.transformer_blocks.0.norm1.weight", "output_blocks.7.1.transformer_blocks.0.norm1.bias", "output_blocks.7.1.transformer_blocks.0.norm2.weight", "output_blocks.7.1.transformer_blocks.0.norm2.bias", "output_blocks.7.1.transformer_blocks.0.norm3.weight", "output_blocks.7.1.transformer_blocks.0.norm3.bias", "output_blocks.7.1.proj_out.weight", "output_blocks.7.1.proj_out.bias", "output_blocks.8.0.in_layers.0.weight", "output_blocks.8.0.in_layers.0.bias", "output_blocks.8.0.in_layers.2.weight", "output_blocks.8.0.in_layers.2.bias", "output_blocks.8.0.emb_layers.1.weight", "output_blocks.8.0.emb_layers.1.bias", "output_blocks.8.0.out_layers.0.weight", "output_blocks.8.0.out_layers.0.bias", "output_blocks.8.0.out_layers.3.weight", "output_blocks.8.0.out_layers.3.bias", "output_blocks.8.0.skip_connection.weight", "output_blocks.8.0.skip_connection.bias", "output_blocks.8.1.norm.weight", "output_blocks.8.1.norm.bias", "output_blocks.8.1.proj_in.weight", "output_blocks.8.1.proj_in.bias", "output_blocks.8.1.transformer_blocks.0.attn1.to_q.weight", "output_blocks.8.1.transformer_blocks.0.attn1.to_k.weight", "output_blocks.8.1.transformer_blocks.0.attn1.to_v.weight", "output_blocks.8.1.transformer_blocks.0.attn1.to_out.0.weight", "output_blocks.8.1.transformer_blocks.0.attn1.to_out.0.bias", "output_blocks.8.1.transformer_blocks.0.ff.net.0.proj.weight", "output_blocks.8.1.transformer_blocks.0.ff.net.0.proj.bias", "output_blocks.8.1.transformer_blocks.0.ff.net.2.weight", "output_blocks.8.1.transformer_blocks.0.ff.net.2.bias", "output_blocks.8.1.transformer_blocks.0.attn2.to_q.weight", "output_blocks.8.1.transformer_blocks.0.attn2.to_k.weight", "output_blocks.8.1.transformer_blocks.0.attn2.to_v.weight", "output_blocks.8.1.transformer_blocks.0.attn2.to_out.0.weight", "output_blocks.8.1.transformer_blocks.0.attn2.to_out.0.bias", "output_blocks.8.1.transformer_blocks.0.norm1.weight", "output_blocks.8.1.transformer_blocks.0.norm1.bias", "output_blocks.8.1.transformer_blocks.0.norm2.weight", "output_blocks.8.1.transformer_blocks.0.norm2.bias", "output_blocks.8.1.transformer_blocks.0.norm3.weight", "output_blocks.8.1.transformer_blocks.0.norm3.bias", "output_blocks.8.1.proj_out.weight", "output_blocks.8.1.proj_out.bias", "output_blocks.8.3.conv.weight", "output_blocks.8.3.conv.bias", "output_blocks.9.0.in_layers.0.weight", "output_blocks.9.0.in_layers.0.bias", "output_blocks.9.0.in_layers.2.weight", "output_blocks.9.0.in_layers.2.bias", "output_blocks.9.0.emb_layers.1.weight", "output_blocks.9.0.emb_layers.1.bias", "output_blocks.9.0.out_layers.0.weight", "output_blocks.9.0.out_layers.0.bias", "output_blocks.9.0.out_layers.3.weight", "output_blocks.9.0.out_layers.3.bias", "output_blocks.9.0.skip_connection.weight", "output_blocks.9.0.skip_connection.bias", "output_blocks.9.1.norm.weight", "output_blocks.9.1.norm.bias", "output_blocks.9.1.proj_in.weight", "output_blocks.9.1.proj_in.bias", "output_blocks.9.1.transformer_blocks.0.attn1.to_q.weight", "output_blocks.9.1.transformer_blocks.0.attn1.to_k.weight", "output_blocks.9.1.transformer_blocks.0.attn1.to_v.weight", "output_blocks.9.1.transformer_blocks.0.attn1.to_out.0.weight", "output_blocks.9.1.transformer_blocks.0.attn1.to_out.0.bias", "output_blocks.9.1.transformer_blocks.0.ff.net.0.proj.weight", "output_blocks.9.1.transformer_blocks.0.ff.net.0.proj.bias", "output_blocks.9.1.transformer_blocks.0.ff.net.2.weight", "output_blocks.9.1.transformer_blocks.0.ff.net.2.bias", "output_blocks.9.1.transformer_blocks.0.attn2.to_q.weight", "output_blocks.9.1.transformer_blocks.0.attn2.to_k.weight", "output_blocks.9.1.transformer_blocks.0.attn2.to_v.weight", "output_blocks.9.1.transformer_blocks.0.attn2.to_out.0.weight", "output_blocks.9.1.transformer_blocks.0.attn2.to_out.0.bias", "output_blocks.9.1.transformer_blocks.0.norm1.weight", "output_blocks.9.1.transformer_blocks.0.norm1.bias", "output_blocks.9.1.transformer_blocks.0.norm2.weight", "output_blocks.9.1.transformer_blocks.0.norm2.bias", "output_blocks.9.1.transformer_blocks.0.norm3.weight", "output_blocks.9.1.transformer_blocks.0.norm3.bias", "output_blocks.9.1.proj_out.weight", "output_blocks.9.1.proj_out.bias", "output_blocks.10.0.in_layers.0.weight", "output_blocks.10.0.in_layers.0.bias", "output_blocks.10.0.in_layers.2.weight", "output_blocks.10.0.in_layers.2.bias", "output_blocks.10.0.emb_layers.1.weight", "output_blocks.10.0.emb_layers.1.bias", "output_blocks.10.0.out_layers.0.weight", "output_blocks.10.0.out_layers.0.bias", "output_blocks.10.0.out_layers.3.weight", "output_blocks.10.0.out_layers.3.bias", "output_blocks.10.0.skip_connection.weight", "output_blocks.10.0.skip_connection.bias", "output_blocks.10.1.norm.weight", "output_blocks.10.1.norm.bias", "output_blocks.10.1.proj_in.weight", "output_blocks.10.1.proj_in.bias", "output_blocks.10.1.transformer_blocks.0.attn1.to_q.weight", "output_blocks.10.1.transformer_blocks.0.attn1.to_k.weight", "output_blocks.10.1.transformer_blocks.0.attn1.to_v.weight", "output_blocks.10.1.transformer_blocks.0.attn1.to_out.0.weight", "output_blocks.10.1.transformer_blocks.0.attn1.to_out.0.bias", "output_blocks.10.1.transformer_blocks.0.ff.net.0.proj.weight", "output_blocks.10.1.transformer_blocks.0.ff.net.0.proj.bias", "output_blocks.10.1.transformer_blocks.0.ff.net.2.weight", "output_blocks.10.1.transformer_blocks.0.ff.net.2.bias", "output_blocks.10.1.transformer_blocks.0.attn2.to_q.weight", "output_blocks.10.1.transformer_blocks.0.attn2.to_k.weight", "output_blocks.10.1.transformer_blocks.0.attn2.to_v.weight", "output_blocks.10.1.transformer_blocks.0.attn2.to_out.0.weight", "output_blocks.10.1.transformer_blocks.0.attn2.to_out.0.bias", "output_blocks.10.1.transformer_blocks.0.norm1.weight", "output_blocks.10.1.transformer_blocks.0.norm1.bias", "output_blocks.10.1.transformer_blocks.0.norm2.weight", "output_blocks.10.1.transformer_blocks.0.norm2.bias", "output_blocks.10.1.transformer_blocks.0.norm3.weight", "output_blocks.10.1.transformer_blocks.0.norm3.bias", "output_blocks.10.1.proj_out.weight", "output_blocks.10.1.proj_out.bias", "output_blocks.11.0.in_layers.0.weight", "output_blocks.11.0.in_layers.0.bias", "output_blocks.11.0.in_layers.2.weight", "output_blocks.11.0.in_layers.2.bias", "output_blocks.11.0.emb_layers.1.weight", "output_blocks.11.0.emb_layers.1.bias", "output_blocks.11.0.out_layers.0.weight", "output_blocks.11.0.out_layers.0.bias", "output_blocks.11.0.out_layers.3.weight", "output_blocks.11.0.out_layers.3.bias", "output_blocks.11.0.skip_connection.weight", "output_blocks.11.0.skip_connection.bias", "output_blocks.11.1.norm.weight", "output_blocks.11.1.norm.bias", "output_blocks.11.1.proj_in.weight", "output_blocks.11.1.proj_in.bias", "output_blocks.11.1.transformer_blocks.0.attn1.to_q.weight", "output_blocks.11.1.transformer_blocks.0.attn1.to_k.weight", "output_blocks.11.1.transformer_blocks.0.attn1.to_v.weight", "output_blocks.11.1.transformer_blocks.0.attn1.to_out.0.weight", "output_blocks.11.1.transformer_blocks.0.attn1.to_out.0.bias", "output_blocks.11.1.transformer_blocks.0.ff.net.0.proj.weight", "output_blocks.11.1.transformer_blocks.0.ff.net.0.proj.bias", "output_blocks.11.1.transformer_blocks.0.ff.net.2.weight", "output_blocks.11.1.transformer_blocks.0.ff.net.2.bias", "output_blocks.11.1.transformer_blocks.0.attn2.to_q.weight", "output_blocks.11.1.transformer_blocks.0.attn2.to_k.weight", "output_blocks.11.1.transformer_blocks.0.attn2.to_v.weight", "output_blocks.11.1.transformer_blocks.0.attn2.to_out.0.weight", "output_blocks.11.1.transformer_blocks.0.attn2.to_out.0.bias", "output_blocks.11.1.transformer_blocks.0.norm1.weight", "output_blocks.11.1.transformer_blocks.0.norm1.bias", "output_blocks.11.1.transformer_blocks.0.norm2.weight", "output_blocks.11.1.transformer_blocks.0.norm2.bias", "output_blocks.11.1.transformer_blocks.0.norm3.weight", "output_blocks.11.1.transformer_blocks.0.norm3.bias", "output_blocks.11.1.proj_out.weight", "output_blocks.11.1.proj_out.bias", "out.0.weight", "out.0.bias", "out.2.weight", "out.2.bias"]
|
models/v2-1_512-ema-pruned.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:88ecb782561455673c4b78d05093494b9c539fc6bfc08f3a9a4a0dd7b0b10f36
|
| 3 |
+
size 5214865159
|
scheduler/scheduler_config.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_class_name": "DDIMScheduler",
|
| 3 |
+
"_diffusers_version": "0.26.1",
|
| 4 |
+
"beta_end": 0.02,
|
| 5 |
+
"beta_schedule": "squaredcos_cap_v2",
|
| 6 |
+
"beta_start": 0.0001,
|
| 7 |
+
"clip_sample": false,
|
| 8 |
+
"clip_sample_range": 1.0,
|
| 9 |
+
"dynamic_thresholding_ratio": 0.995,
|
| 10 |
+
"num_train_timesteps": 1000,
|
| 11 |
+
"prediction_type": "v_prediction",
|
| 12 |
+
"rescale_betas_zero_snr": true,
|
| 13 |
+
"sample_max_value": 1.0,
|
| 14 |
+
"set_alpha_to_one": true,
|
| 15 |
+
"steps_offset": 1,
|
| 16 |
+
"thresholding": false,
|
| 17 |
+
"timestep_spacing": "leading",
|
| 18 |
+
"trained_betas": null
|
| 19 |
+
}
|
source/VGen.jpg
ADDED
|
source/fig_vs_vgen.jpg
ADDED
|
source/i2vgen_fig_01.jpg
ADDED
|
source/i2vgen_fig_02.jpg
ADDED
|
source/i2vgen_fig_04.png
ADDED
|
Git LFS Details
|
source/logo.png
ADDED
|
text_encoder/config.json
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "i2vgen-xl/text_encoder",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"CLIPTextModel"
|
| 5 |
+
],
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"bos_token_id": 0,
|
| 8 |
+
"dropout": 0.0,
|
| 9 |
+
"eos_token_id": 2,
|
| 10 |
+
"hidden_act": "gelu",
|
| 11 |
+
"hidden_size": 1024,
|
| 12 |
+
"initializer_factor": 1.0,
|
| 13 |
+
"initializer_range": 0.02,
|
| 14 |
+
"intermediate_size": 4096,
|
| 15 |
+
"layer_norm_eps": 1e-05,
|
| 16 |
+
"max_position_embeddings": 77,
|
| 17 |
+
"model_type": "clip_text_model",
|
| 18 |
+
"num_attention_heads": 16,
|
| 19 |
+
"num_hidden_layers": 24,
|
| 20 |
+
"pad_token_id": 1,
|
| 21 |
+
"projection_dim": 1024,
|
| 22 |
+
"torch_dtype": "float16",
|
| 23 |
+
"transformers_version": "4.36.2",
|
| 24 |
+
"vocab_size": 49408
|
| 25 |
+
}
|
text_encoder/model.fp16.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0eb4cf70d7768f4f61bb25e98b2ddaf545f6b33ecc2d7cc3eaa4670a09722fd2
|
| 3 |
+
size 706014768
|
text_encoder/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bd94a7ea6922e8028227567fe14e04d2989eec31c482e0813e9006afea6637f1
|
| 3 |
+
size 1411983168
|
tokenizer/merges.txt
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer/special_tokens_map.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token": {
|
| 3 |
+
"content": "<|startoftext|>",
|
| 4 |
+
"lstrip": false,
|
| 5 |
+
"normalized": true,
|
| 6 |
+
"rstrip": false,
|
| 7 |
+
"single_word": false
|
| 8 |
+
},
|
| 9 |
+
"eos_token": {
|
| 10 |
+
"content": "<|endoftext|>",
|
| 11 |
+
"lstrip": false,
|
| 12 |
+
"normalized": false,
|
| 13 |
+
"rstrip": false,
|
| 14 |
+
"single_word": false
|
| 15 |
+
},
|
| 16 |
+
"pad_token": {
|
| 17 |
+
"content": "<|endoftext|>",
|
| 18 |
+
"lstrip": false,
|
| 19 |
+
"normalized": false,
|
| 20 |
+
"rstrip": false,
|
| 21 |
+
"single_word": false
|
| 22 |
+
},
|
| 23 |
+
"unk_token": {
|
| 24 |
+
"content": "<|endoftext|>",
|
| 25 |
+
"lstrip": false,
|
| 26 |
+
"normalized": false,
|
| 27 |
+
"rstrip": false,
|
| 28 |
+
"single_word": false
|
| 29 |
+
}
|
| 30 |
+
}
|
tokenizer/tokenizer_config.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"added_tokens_decoder": {
|
| 4 |
+
"49406": {
|
| 5 |
+
"content": "<|startoftext|>",
|
| 6 |
+
"lstrip": false,
|
| 7 |
+
"normalized": true,
|
| 8 |
+
"rstrip": false,
|
| 9 |
+
"single_word": false,
|
| 10 |
+
"special": true
|
| 11 |
+
},
|
| 12 |
+
"49407": {
|
| 13 |
+
"content": "<|endoftext|>",
|
| 14 |
+
"lstrip": false,
|
| 15 |
+
"normalized": false,
|
| 16 |
+
"rstrip": false,
|
| 17 |
+
"single_word": false,
|
| 18 |
+
"special": true
|
| 19 |
+
}
|
| 20 |
+
},
|
| 21 |
+
"bos_token": "<|startoftext|>",
|
| 22 |
+
"clean_up_tokenization_spaces": true,
|
| 23 |
+
"do_lower_case": true,
|
| 24 |
+
"eos_token": "<|endoftext|>",
|
| 25 |
+
"errors": "replace",
|
| 26 |
+
"model_max_length": 77,
|
| 27 |
+
"pad_token": "<|endoftext|>",
|
| 28 |
+
"tokenizer_class": "CLIPTokenizer",
|
| 29 |
+
"unk_token": "<|endoftext|>"
|
| 30 |
+
}
|
tokenizer/vocab.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
unet/config.json
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_class_name": "I2VGenXLUNet",
|
| 3 |
+
"_diffusers_version": "0.26.1",
|
| 4 |
+
"_name_or_path": "i2vgen-xl/unet",
|
| 5 |
+
"attention_head_dim": 64,
|
| 6 |
+
"block_out_channels": [
|
| 7 |
+
320,
|
| 8 |
+
640,
|
| 9 |
+
1280,
|
| 10 |
+
1280
|
| 11 |
+
],
|
| 12 |
+
"cross_attention_dim": 1024,
|
| 13 |
+
"down_block_types": [
|
| 14 |
+
"CrossAttnDownBlock3D",
|
| 15 |
+
"CrossAttnDownBlock3D",
|
| 16 |
+
"CrossAttnDownBlock3D",
|
| 17 |
+
"DownBlock3D"
|
| 18 |
+
],
|
| 19 |
+
"in_channels": 4,
|
| 20 |
+
"layers_per_block": 2,
|
| 21 |
+
"norm_num_groups": 32,
|
| 22 |
+
"num_attention_heads": 64,
|
| 23 |
+
"out_channels": 4,
|
| 24 |
+
"sample_size": 32,
|
| 25 |
+
"up_block_types": [
|
| 26 |
+
"UpBlock3D",
|
| 27 |
+
"CrossAttnUpBlock3D",
|
| 28 |
+
"CrossAttnUpBlock3D",
|
| 29 |
+
"CrossAttnUpBlock3D"
|
| 30 |
+
]
|
| 31 |
+
}
|
unet/diffusion_pytorch_model.fp16.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6ef043141380ab4f5f0698d0b735e4020d6049819e3c756ddbe2672a48a466d4
|
| 3 |
+
size 2841124432
|
unet/diffusion_pytorch_model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3e7c73b3ff159580a1a6535b1ccb473b09a2f40853c03f5d546db70632456ab8
|
| 3 |
+
size 5682063336
|
vae/config.json
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_class_name": "AutoencoderKL",
|
| 3 |
+
"_diffusers_version": "0.26.1",
|
| 4 |
+
"_name_or_path": "i2vgen-xl/vae",
|
| 5 |
+
"act_fn": "silu",
|
| 6 |
+
"block_out_channels": [
|
| 7 |
+
128,
|
| 8 |
+
256,
|
| 9 |
+
512,
|
| 10 |
+
512
|
| 11 |
+
],
|
| 12 |
+
"down_block_types": [
|
| 13 |
+
"DownEncoderBlock2D",
|
| 14 |
+
"DownEncoderBlock2D",
|
| 15 |
+
"DownEncoderBlock2D",
|
| 16 |
+
"DownEncoderBlock2D"
|
| 17 |
+
],
|
| 18 |
+
"force_upcast": true,
|
| 19 |
+
"in_channels": 3,
|
| 20 |
+
"latent_channels": 4,
|
| 21 |
+
"layers_per_block": 2,
|
| 22 |
+
"norm_num_groups": 32,
|
| 23 |
+
"out_channels": 3,
|
| 24 |
+
"sample_size": 768,
|
| 25 |
+
"scaling_factor": 0.18125,
|
| 26 |
+
"up_block_types": [
|
| 27 |
+
"UpDecoderBlock2D",
|
| 28 |
+
"UpDecoderBlock2D",
|
| 29 |
+
"UpDecoderBlock2D",
|
| 30 |
+
"UpDecoderBlock2D"
|
| 31 |
+
]
|
| 32 |
+
}
|
vae/diffusion_pytorch_model.fp16.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3e4c08995484ee61270175e9e7a072b66a6e4eeb5f0c266667fe1f45b90daf9a
|
| 3 |
+
size 167335342
|
vae/diffusion_pytorch_model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2aa1f43011b553a4cba7f37456465cdbd48aab7b54b9348b890e8058ea7683ec
|
| 3 |
+
size 334643268
|