Stanford's BulletTime: decoupling frame index from world time in video generation
GordonWetzstein · x · 2026-09-09
Gordon Wetzstein shares a key insight from BulletTime: frame index and world time should be separate. Explicit time and camera controls let the model freeze the action while moving the camera; simply retiming inputs before changing viewpoints loses information and consistency.
Related event: Stanford's BulletTime Decouples Frame Index from World Time(2 posts)→
More from Multimodal
- One-Take Lava Apocalypse Video Generated with Seedance 2.5 — LudovicCreator · 2026-09-09
- Reimagining Hitchcock's Psycho with Blender and GPT-6 Astra: a 27-cut shower scene study — taherdhanera · 2026-09-09
- Plan the shot in 3D first, then let Seedance 2.5 generate the video — thetripathi58 · 2026-09-09
- Mirelo SFX 1.6 fixes Seedance 2.5 audio without regenerating the whole video — ordax · 2026-09-09
- Wan Vace workflow: 5 keyframes in one video with per-frame control — o0ANARKY0o · 2026-09-09
- AI Lip-Sync Could Be the Missing Link in Video Translation, Tests of SyncSo, Rask and HeyGen Show — NewPhoneWhotiz · 2026-09-09