<img src="assets/dragdiffusion.png" width="750" alt="Architecture of DragDiffuison"/>

DragDiffusion

Unofficial implementation of DragDiffusion: Harnessing Diffusion Models for Interactive Point-based Image Editing.

# requirements
conda env create -f environment.yml
conda activate DragDiffusion
pip install -r requirements.txt
# To obtain the feature from StableDiffusion Unet (WIP)
mv assets/unet_2d_condition.py YOUR_CONDA_ENV/site-packages/diffusers/models/
# run demo
python visualizer_drag_gradio.py

TODO

• [x] drag process
• [x] mask
• [x] Gradio GUI
• [ ] imgui GUI

StableDiffusion Pre-Trained Model

Following Diffusers to obtain pre-trained StableDiffuison model.

References

• https://github.com/rosinality/stylegan2-pytorch
• https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_lora.py