Overworld released Waypoint-1, a real-time interactive video diffusion model controllable via text, mouse, and keyboard. Trained from scratch on 10,000 hours of video game footage, it uses a frame-causal rectified flow transformer with diffusion forcing and self-forcing post-training. The 2.3B Waypoint-1-Small variant runs through Overworld's WorldEngine inference library, achieving 30 FPS at 4 denoising steps on a 5090 GPU. Waypoint-1-Medium is listed as coming soon.
No score is assigned. Sources and their independence are shown in the citation chain below.