Deploy Oasis AI-generated Minecraft Server Tutorial

November 7, 2024

You can deploy your own oasis minecraft server by Oasis 500M model. Oasis is an interactive world model that uses diffusion transformers to generate gameplay based on user input, allowing you to create dynamic, AI-driven game worlds.

oasis minecraft village outpost

What is Oasis?

Oasis is an innovative AI model designed to transform the way games are played. By processing keyboard inputs, it generates gameplay in an autoregressive manner. The Oasis 500M model, a scaled-down version of the original, is now available for public use, complete with inference code for action-conditional frame generation.

Getting Started

To deploy your own Oasis Minecraft server, you'll first need to set up the necessary environment. Follow these steps to get started:

Step 1: Clone the Repository

Start by cloning the Oasis repository from GitHub:

git clone https://github.com/etched-ai/open-oasis.git
cd open-oasis

Step 2: Install Dependencies

Ensure you have PyTorch installed, along with other required libraries:

# Install PyTorch
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu121

# Install additional dependencies
pip install einops diffusers timm av

Step 3: Download Model Weights

Next, download the model weights from Hugging Face. This step requires you to log in:

huggingface-cli login
huggingface-cli download Etched/oasis-500m oasis500m.safetensors # DiT checkpoint
huggingface-cli download Etched/oasis-500m vit-l-20.safetensors  # ViT VAE checkpoint

Running the Model

With everything set up, you can now generate gameplay frames using the Oasis model. A basic script is provided to help you get started:

python generate.py

This script will load a prompt frame from a video and generate additional frames based on user actions. The output is saved as video.mp4.

Experiment with Different Environments

To fully explore the capabilities of Oasis, try swapping out the .mp4 input file in the script. This allows you to test the model in various gaming environments, providing a unique and personalized experience each time.

For more information and to try the most powerful version of the model, visit the live demo and check out the source code.