Waypoint-1.5: AI generates interactive 720p worlds on an ordinary RTX 3090
Why it matters
Overworld has released Waypoint-1.5, a real-time video world model that generates interactive virtual environments at 720p/60fps on consumer GPUs such as the RTX 3090 and Apple Silicon Macs. The model was trained on ~100x more data than the previous version and is available open source on HuggingFace.
On April 9, Overworld unveiled Waypoint-1.5, a new generation of its real-time video world model β a generative AI system that creates virtual worlds in real time that users can explore and interact with. The key difference from its predecessors is the focus on accessibility on consumer hardware instead of expensive datacenter GPUs.
Technical specifications
The model comes in two sizes: Waypoint-1.5-1B (1 billion parameters, 720p resolution for high-end PCs) and a 360P variant for wider hardware coverage. Performance: 720p @ 60 fps on RTX 3090-5090, gaming laptops and β soon β Apple Silicon Mac devices. This is a significant leap from the previous version, where similar capabilities required datacenter compute.
Training: approximately 100x more data than Waypoint-1, which dramatically improved frame-to-frame coherence and motion consistency. The team has also developed more efficient video modeling techniques that reduce redundant computation between adjacent frames.
How to access it
The model is available on Hugging Face (Overworld/Waypoint-1.5-1B and Waypoint-1.5-1B-360P). For local execution there is Overworld Biome, an open-source desktop client on GitHub, as well as World Engine β the core inference library around which roughly a dozen third-party clients have already emerged. For those who prefer not to install anything, there is also instant browser access via overworld.stream.
What this changes
Until now, generative world models have mostly been an academic curiosity or the exclusive domain of large labs. Waypoint-1.5 lowers the minimum spec to the level of an average gaming PC, opening the door to interactive entertainment, creative tools, simulations and AI-native research environments. The open license and the ecosystem of community tools around it suggest that this could be for 3D world models what Stable Diffusion was for 2D image generation.