A repository that uses the Diffusers library to finetune and train Stable Diffusion models in PyTorch Lightning. Written for experiments surrounding permissive datasets and data augmentation. Still a work in progress.
- Train or finetune a Diffuser U-Net.
- Train or finetune a VAE prior.
- Train or finetune a CLIP encoder as the LDM's text encoder.
- Create preprocessor and dataloader for Unsplash-25K.
- Train LDM on Unsplash-25K dataset.
- Train LDM+CLIP on Unsplash-25K dataset.