Simulate Anything on Anything - Open science world models that run on consumer hardware. From mobile phones to gaming PCs, experience state-of-the-art AI models at blazing speeds.

Join Our Discord | Visit Our Website | Try OWL Eval

We're making world models accessible to everyone. Through open science and community collaboration, we're building AI models that run locally on your hardware - no cloud required. Our models achieve hundreds of FPS on RTX 5090s and 60 FPS on NVIDIA Jetson for robotics applications.

• 60 FPS on the latest consumer hardware
• No cloud computing needed. Run our best models on your current PC or laptop

• Production-ready optimization techniques

• OWL Eval - An open-source evaluation platform for AI-generated videos
• Reveals critical model failures that automated metrics miss
• Understand human perception of AI content

Depth map integration and point cloud conditioning for spatially coherent world generation.

Solving temporal flickering through frame-pair training and optical flow conditioning.

Diffusion transformers for decoding, hierarchical VAE-GANs, and self-forcing for KV cache optimization.

Models intended to run anywhere from Steam Deck to iPhone (soon), achieving real-time performance without cloud infrastructure.

We offer paid bounties for contributions to our open-source projects. Check our GitHub Bounty Board for current opportunities.

1. Browse active bounties on our bounty board
2. Join our Discord (1000+ members) or our model-specific Discord (*Ask for a role!)
3. Contact @louis or @spencer on Discord to discuss or claim a bounty
4. Submit your PR and get paid upon merge

• World Models - State-of-the-art models for real-time generation
• OWL Eval - Human evaluation platform for AI-generated content
• Diffusion Decoders - Optimized decoders with depth pruning
• Training Pipelines - Open frameworks for model training

• Why Human Evaluation is the Missing Piece in World Model Development
• Optimizing Diffusion Decoders with Depth Pruning
• Generation vs. Reconstruction: Striking A Balance
• Fast Audio Video World Models
• Training Diffusable Autoencoders

• 🎮 Discord (1000+ researchers/developers) / model-specific & consumer facing Discord (*Ask for a role!)
• 💻 GitHub Organization
• 🌐 Website
• 🧪 Evaluation Platform
• 🐦 Twitter/X

• Researchers - Collaborate on cutting-edge world model research
• Developers - Build applications with our models or contribute code
• Roboticists - Deploy models on embedded systems for real-time inference
• Investors - Learn about investment opportunities in open science AI
• Partners - Explore collaboration and integration possibilities

This project is licensed under the MIT License - see the LICENSE file for details.

Built by the OWL team in Providence, Rhode Island