kerzgrr/diffusionbreakout-base

Diffusion Breakout Base

The unconditional Breakout world model used as the starting point for the playable version.

128x128 at 6 FPS with 12 frames of latent history. The model has about 70M parameters and uses the frozen SDXL VAE from madebyollin/sdxl-vae-fp16-fix.

Run it

Requires a CUDA GPU with BF16 support.

pip install torch numpy pillow safetensors huggingface_hub diffusers

hf download kerzgrr/diffusionbreakout-base live_infer.py --local-dir .
python live_infer.py --steps 1

To download everything first:

hf download kerzgrr/diffusionbreakout-base \
  --local-dir checkpoints/diffusionbreakout-base \
  --include "ema.safetensors" \
  --include "config.json" \
  --include "live_infer.py"

python checkpoints/diffusionbreakout-base/live_infer.py \
  --local-dir checkpoints/diffusionbreakout-base \
  --steps 1 \
  --window-scale 6

This model is unconditional, so there are no paddle controls. It rolls forward from a fresh game context.

Files

file	contents
ema.safetensors	unconditional EMA weights
config.json	model, video, codec, and training settings
live_infer.py	single-file live inference
preview.png	validation strip