Constant starting noise for demo() images

crowsonkb / k-diffusion

Karras et al. (2022) diffusion models for PyTorch

MIT License

2.26k stars 372 forks source link

Constant starting noise for demo() images #28

Open bob80333 opened 1 year ago

bob80333 commented 1 year ago

Keep the starting noise for sampled image grids constant across all training steps. This makes comparing images generated across training easier, and also allows for creating training montages.