[HUGGINGFACE]score: 0.45
minWM Open-Source Framework Converts Video Diffusion Models into Real-Time World Models
May 27, 2026
minWM provides an end-to-end open-source pipeline — data construction, controllable fine-tuning, autoregressive training, few-step distillation, and streaming inference — to convert bidirectional T2V/TI2V models into camera-controllable real-time world models. It targets the full stack required for low-latency interactive rollout, which existing video diffusion models don't support natively.
paper
HOW THIS AFFECTS YOU
●
builderYou can use minWM today to convert an existing video diffusion model into a real-time interactive world model without building the distillation and streaming inference pipeline from scratch.
●
researcherThe full-stack open-source release makes it easier to benchmark and iterate on interactive world model architectures end-to-end.