Why video models break physics: trajectory locked in first 10% of denoising
linoy_tsaban · x · 2026-09-25
Arguing that video models are now convincing enough that we should scrutinize their physical faithfulness, the author highlights a mechanistic interpretability study: an object's trajectory gets locked in during the first 10% of denoising, and RoPE's spatial decay locks it onto the wrong position. A one-line change to the rotary embedding fixes it. Since the paper only tested Wan2.1-1.3B, she is now applying the fix to H3 to see if it generalizes.
More from Multimodal
- MiniMax H3-Image image model debuts on Tencent's Miora, free to try Nov 11–17 — aziz4ai · 2026-11-11
- MiniMax H3 hits 100 days open-source: 1,472 derivative models, 148M+ downloads — linoy_tsaban · 2026-11-11
- MiniMax H3-Image explodes an espresso machine photo into 15 labeled parts — Charlietest123 · 2026-11-11
- Indie dev built a tool that clips best moments from 4-hour streams in 30 minutes — kissaadams · 2026-11-11
- Non-Blender user makes a video in 30 minutes with Claude and Seedance — FinanceYF5 · 2026-11-11
- Whimsical Woodland: A Watercolor AI Art Prompt Template for Dreamy Forest Scenes — LudovicCreator · 2026-11-11