FateZero: Fusing Attentions for Zero-shot Text-based Video Editing

Chenyang Qi, Xiaodong Cun, Yong Zhang, Chenyang Lei, Xintao Wang, Ying Shan, and Qifeng Chen

Paper | Project Page | Code

Abstract

The diffusion-based generative models have achieved remarkable success in text-based image generation. However, since it contains enormous randomness in generation progress, it is still challenging to apply such models for real-world visual content editing, especially in videos. In this paper, we propose FateZero, a zero-shot text-based editing method on real-world videos without per-prompt training or use-specific mask. To edit videos consistently, we propose several techniques based on the pre-trained models. Firstly, in contrast to the straightforward DDIM inversion technique, our approach captures intermediate attention maps during inversion, which effectively retain both structural and motion information. These maps are directly fused in the editing process rather than generated during denoising. To further minimize semantic leakage of the source video, we then fuse self-attentions with a blending mask obtained by cross-attention features from the source prompt. Furthermore, we have implemented a reform of the self-attention mechanism in denoising UNet by introducing spatial-temporal attention to ensure frame consistency. Yet succinct, our method is the first one to show the ability of zero-shot text-driven video style and local attribute editing from the trained text-to-image model. We also have a better zero-shot shape-aware editing ability based on the text-tovideo model. Extensive experiments demonstrate our superior temporal consistency and editing capability than previous works.

Changelog

• 2023.03.17 Release Code and Paper!

Todo

• Release the edit config for teaser
• Memory and runtime profiling and tune-a-video optimization
• More detailed description of our environment and More Hands-on guidance
• Release configs for other result and in-the-wild dataset
• Release more application

Setup Environment

Our method is tested using cuda11, fp16 and xformers on a single A100 or 3090.

conda create -n fatezero38 python=3.8
conda activate fatezero38

pip install -r requirements.txt

xformers is recommended for A100 GPU to save memory and running time. We find its installation not stable. You may try the following wheel:

wget https://github.com/ShivamShrirao/xformers-wheels/releases/download/4c06c79/xformers-0.0.15.dev0+4c06c79.d20221201-cp38-cp38-linux_x86_64.whl
pip install xformers-0.0.15.dev0+4c06c79.d20221201-cp38-cp38-linux_x86_64.whl

Validate the installation by

python test_install.py

Our environment is similar to Tune-A-video (official, unofficial) and prompt-to-prompt. You may check them for more details.

FateZero Editing

Download the stable diffusion v1-4 (or other interesting image diffusion model) and put it to ./ckpt/stable-diffusion-v1-4. You may refer to the following bash command:

mkdir ./ckpt
# download from huggingface face, takes 20G space
git lfs install
git clone https://huggingface.co/CompVis/stable-diffusion-v1-4
cd ./ckpt
ln -s ../stable-diffusion-v1-4 .

The directory structure should like this:

ckpt
├── stable-diffusion-v1-4
├── stable-diffusion-v1-5
...
data
├── car-turn
│   ├── 00000000.png
│   ├── 00000001.png
│   ├── ...
video_diffusion

You could generate style editing result in our teaser by running:

CUDA_VISIBLE_DEVICES=0 python test_fatezero.py --config config/teaser/jeep_watercolor.yaml

The result is saved as follows:

result
├── teaser
│   ├── jeep_watercolor
│           ├── cross-attention
│           ├── sample
│           ├── train_samples

where cross-attention is the visualization of cross-attention during inversion; sample is the result videos obtained from target prompt; train_sample is the input video;

We also provide a Tune-A-Video checkpoint. You may download the check point is this link and move it to ./ckpt/jeep_tuned_200/. Run following command to get the result:

CUDA_VISIBLE_DEVICES=0 python test_fatezero.py --config config/teaser/jeep_posche.yaml

Results

Please check our Project Page for more video editing results.

Demo Video

165a65fe9b83096a92a1bddb9bfff459.mp4

Citation

@misc{qi2023fatezero,
      title={FateZero: Fusing Attentions for Zero-shot Text-based Video Editing}, 
      author={Chenyang Qi and Xiaodong Cun and Yong Zhang and Chenyang Lei and Xintao Wang and Ying Shan and Qifeng Chen},
      year={2023},
      eprint={2303.09535},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Acknowledgements

This repository borrows heavily from Tune-A-Video and prompt-to-prompt. thanks the authors for sharing their code and models.

Maintenance

This is the codebase for our research work. We are still working hard to update this repo and more details are coming in days. If you have any questions or ideas to discuss, feel free to contact Chenyang Qi or Xiaodong Cun.