Cleaned up the code from Stable UnCLIP 2.1 removing a lot of the bells and whistle from the original to improve readability, Implemented stricter GPU memory management allowing me to run inpainting of 768x768 image on a 6GB GPU.
Usage: download the checkpoints for inpainting and put them in the checkpoints directory
python LatentDiffusionInpainting.py <path-to-checkpoint>
I'll be working on a docker container to run this on AWS, but the roadmap for when i come back includes:
- Implementation of the other gradio functions
- Allowing for swapping of modules and finetuning
let me know if you think other directions might be interesting
⭐ if you find useful so i know to release more tools in this direction