GameNgen fine-tuning - arnaudstiegler/gameNgen-baseline-20ksteps

Full finetune of CompVis/stable-diffusion-v1-4. The weights were fine-tuned on the P-H-B-D-a16z/ViZDoom-Deathmatch-PPO-Lrg dataset. You can find some example images in the following.

Intended uses & limitations

How to use

See: https://github.com/arnaudstiegler/gameNgen-repro/blob/main/README.md

Training details

20k steps batch size 12 10 epochs with CFG data augmentation