lightning-diffusion

Getting Started

Help

python scripts/main.py fit --help

Make config

python scripts/main.py fit --model=StableDiffusionModule --data=HFDataModule --print_config

Train

python scripts/main.py fit -c 'your config path'

Pre-compute embs

python scripts/precompute.py -c config/text_to_image/stable_diffusion.yaml --wds_save_dir=../data_cache --repeat=10

TODO

• high priority
  • dreambooth
  • lcm, (dmd, pcm)
• other
  • upscale
  • stable cascade
  • stable diffusion 3
  • HDiT
  • t2i-adapter
  • textual-inversion
  • deepfloyd
  • prompt-free diffusion
  • esd
  • aspect ratio bucketing
  • save part of checkpoint
  • tips from kohya-ss/sd-scripts
  • video, 3d, ... generation

Acknowledgement

This repo borrows the architecture design and part of the code from diffengine, diffusers, naifu, generative-models, sd-scripts.