Skip to content

Latest commit

 

History

History
23 lines (20 loc) · 857 Bytes

README.md

File metadata and controls

23 lines (20 loc) · 857 Bytes

Shahbuland Generative Models (SGM for short)

General repo for experiments/learnings to work towards dynamic AI generated video games (holo-deck or bust)

TODO

  • Note: I don't plan on doing all these things in depth in this repository but I consider them to be preliminaries for AI generated games to be feasible
  • MM-DiT for videos
  • rectified flow mm transformer for videos
  • Latent video diffusion (SVD VAE?)
  • V prediction for diffusion
  • Smarter noise schedule
  • Flow matching/rectified flows
  • Final frame inpainting
  • QKV caching for faster inpainting
  • Flash Attention
  • Adversarial latent diffusion distillation for video models
  • VAE training
  • Adversarial VAE decoding
  • VAE distillation
  • Video VAE
  • Control prediction
  • Control conditioned video generation