README.md

June 18, 2025 ยท View on GitHub

gazelle cat frame_extrapolation fish_base candle surfer-0 flower_v2 horse-2 squirrel pixar_i

๐Ÿค— HuggingFace Demo

Try out our Hugging Face Space!

Powered by ZeroGPU - Special thanks to Hugging Face ๐Ÿ™Œ

๐Ÿš€ QuickStart

  • Install conda environment:
conda env create -f flipsketch.yml
  • Download T2V LoRA Model from HuggingFace
git lfs install
git clone https://huggingface.co/Hmrishav/t2v_sketch-lora
  • Place LoRA checkpoint under root folder:
mv t2v_sketch-lora/checkpoint-2500 ./checkpoint-2500/
  • Run app
python app.py

โšก PyTorch 2+ Support:

To use the codebase with PyTorch2.0, modify here to import from text2vid_torch2.py instead of text2vid_modded.py

๐Ÿ’ก How it works?

Generation Pipeline

We use a T2V model fine-tuned on sketch animations, and condition it to follow an input sketch.

Noise Refinement

We perform attention composition with reference noise from the input sketch.

๐Ÿ™๐Ÿป Acknowledgements

๐Ÿ“’ Citation

If you find FlipSketch useful, consider citing our work:

@inproceedings{bandyopadhyay2024flipsketch,
  title={FlipSketch: Flipping static Drawings to Text-Guided Sketch Animations}, 
  author={Hmrishav Bandyopadhyay and Yi-Zhe Song},
  booktitle={CVPR},
  year={2025}
}