Oasis is an interactive world model developed by Decart and Etched. Based on diffusion transformers, Oasis takes in user keyboard input and generates gameplay in an autoregressive manner. We release the weights for Oasis 500M, a downscaled version of the model, along with inference code for action-conditional frame generation.
For more details, see our joint blog post to learn more.
git clone https://github.com/etched-ai/open-oasis.git
cd open-oasis
pip install -r requirements.txt
huggingface-cli login
huggingface-cli download Etched/oasis-500m oasis500m.pt # DiT checkpoint
huggingface-cli download Etched/oasis-500m vit-l-20.pt # ViT VAE checkpoint
We include a basic inference script that loads a prompt frame from a video and generates additional frames conditioned on actions.
python inference.py
The resulting video will be saved to video.mp4.