Text2Sim turns text prompts into editable physics-grounded videos and sim setups

erwincoumans · x · 2026-10-09

Minchen Li's team released Text2Sim, praised by Bullet physics creator Erwin Coumans. The insight: video generation looks nice but is hard to edit, especially when physics is involved. Text2Sim converts a text prompt into a full sim + render setup, producing physically grounded video you can actually edit — plus unlimited training data for video models and VLMs. Coumans notes he first tried text-to-sim back in 2023 when ChatGPT launched, and simulators and foundation models have advanced enormously since.

Original post →

More from Embodied

Embodied channel →