minWM: A Full-Stack Open-Source Framework for Real-Time Interactive Video World Models

Researchers released minWM, an open-source framework for turning bidirectional video diffusion models into interactive world models.

It fine-tunes text-to-video and image-to-video backbones, then applies causal training and distillation to enable few-step autoregressive rollout.

The project ships runnable code, checkpoints, docs, and experiments across open backbones and existing world models.

By lowering latency and improving camera control, minWM could make real-time video simulation and generation more practical.