StyleCrafter: Enhancing Stylized Text-to-Video Generation with Style Adapter

🔥🔥🔥 StyleCrafter on SDXL for stylized image generation is available! Enabling higher resolution(1024×1024) and more visually pleasing!

                 
            _**[GongyeLiu](https://github.com/GongyeLiu), [Menghan Xia*](https://menghanxia.github.io/), [Yong Zhang](https://yzhang2016.github.io), [Haoxin Chen](https://scholar.google.com/citations?user=6UPJSvwAAAAJ&hl=zh-CN&oi=ao), [Jinbo Xing](https://doubiiu.github.io/),
[Xintao Wang](https://xinntao.github.io/), [Yujiu Yang*](https://scholar.google.com/citations?user=4gH3sxsAAAAJ&hl=zh-CN&oi=ao), [Ying Shan](https://scholar.google.com/citations?hl=en&user=4oXBp9UAAAAJ&view_op=list_works&sortby=pubdate)**_

(* corresponding authors) From Tsinghua University and Tencent AI Lab.

🔆 Introduction

TL;DR: We propose StyleCrafter, a generic method that enhances pre-trained T2V models with style control, supporting Style-Guided Text-to-Image Generation and Style-Guided Text-to-Video Generation.

1. ⭐⭐ Style-Guided Text-to-Video Generation.

Style-guided text-to-video results. Resolution: 320 x 512; Frames: 16. (Compressed)

2. Style-Guided Text-to-Image Generation.

Style-guided text-to-image results. Resolution: 512 x 512. (Compressed)

📝 Changelog

• [2024.07.14]: 🔥🔥 Remove watermarks! We finetune temporal blocks on non-watermark data with 500 steps, download updated checkpoints here.
• [2024.06.25]: 🔥🔥 Support StyleCrafter on SDXL!
• [2023.12.08]: 🔥🔥 Release the Huggingface online demo.
• [2023.12.05]: 🔥🔥 Release the code and checkpoint.
• [2023.11.30]: 🔥🔥 Release the project page.

🧰 Models

It takes approximately 5 seconds to generate a 512×512 image and 85 seconds to generate a 320×512 video with 16 frames using a single NVIDIA A100 (40G) GPU. A GPU with at least 16G GPU memory is required to perform the inference process.

⚙️ Setup

conda create -n stylecrafter python=3.8.5
conda activate stylecrafter
pip install -r requirements.txt

💫 Inference

1) Download all checkpoints according to the instructions 2) Run the commands in terminal.

# style-guided text-to-image generation
sh scripts/run_infer_image.sh

# style-guided text-to-video generation
sh scripts/run_infer_video.sh

3) (Optional) Infernce on your own data according to the instructions

👨‍👩‍👧‍👦 Crafter Family

VideoCrafter1: Framework for high-quality text-to-video generation.

ScaleCrafter: Tuning-free method for high-resolution image/video generation.

TaleCrafter: An interactive story visualization tool that supports multiple characters.

LongerCrafter: Tuning-free method for longer high-quality video generation.

DynamiCrafter Animate open-domain still images to high-quality videos.

📢 Disclaimer

We develop this repository for RESEARCH purposes, so it can only be used for personal/research/non-commercial purposes.

🙏 Acknowledgements

We would like to thank AK(@_akhaliq) for the help of setting up online demo.

📭 Contact

If your have any comments or questions, feel free to contact lgy22@mails.tsinghua.edu.cn

BibTex

@article{liu2023stylecrafter,
  title={StyleCrafter: Enhancing Stylized Text-to-Video Generation with Style Adapter},
  author={Liu, Gongye and Xia, Menghan and Zhang, Yong and Chen, Haoxin and Xing, Jinbo and Wang, Xintao and Yang, Yujiu and Shan, Ying},
  journal={arXiv preprint arXiv:2312.00330},
  year={2023}
}