NVIDIA's WorldTrace fixes scene forgetting in video world models, NeurIPS 2026 paper
cindy_x_wu · x · 2026-11-21
In a NeurIPS 2026 paper, NVIDIA researchers introduce WorldTrace, addressing a core flaw of video world models: generated interactive worlds forget visited places—turn the camera away and back, and a garage grows a third door while the garden and roof vanish. WorldTrace keeps scenes consistent when revisited at the same cache size, versus the memory loss of standard sliding-window caches.
More from Multimodal
- Creator shows HeyGen-powered video edit: 'editing to another level' — aziz4ai · 2026-11-21
- AI-built oil painting engine: 2,300 commits in 8 days as Claude Code runs a 'studio' of sessions — FotoRe_store · 2026-11-21
- "AGI Before GTA6": AI-Generated Music Video Riffs on 13-Year GTA VI Wait — Yamapama · 2026-11-21
- Meta's AVTR adds visual imagination to multimodal reasoning, +4.5 pts across 16 benchmarks — liuzhuang1234 · 2026-11-21
- Genie 4 world model live on Comfy playground, early testers impressed — EnvironmentalDay7419 · 2026-11-21
- Roblox-style AI horror short built with CUT + local ComfyUI, workflow shared — rileygstaliger · 2026-11-21