Meta unveils AVM: autoregressive view model streams 2D views as you steer a virtual camera
JonathonLuiten · x · 2026-11-20
Meta researchers introduce AVM (Autoregressive View Model): given one or more images/videos of a complex 3D scene, it streams 2D views as you steer a virtual camera, reconstructing what the context supports and imagining the rest — without building an explicit 3D representation. Jonathon Luiten praises its free movement through both time and space, likening it to a better World Labs.
More from Multimodal
- Kitten TTS 2 Mini Trends on Hugging Face: Lightweight Multilingual Open TTS — KittenML · 2026-11-20
- World Snapshot Models unify 1,000+ visual tasks as generation — World-Snapshot · 2026-11-20
- Character LoRA merges pollute art style? Illustrious user asks for a fix — Maximum_Cause4444 · 2026-11-20
- InfiniGen: a Windows local AI image/video studio wrapping ComfyUI, with AI shot planning — methodangel · 2026-11-20
- AI-generated Harry Potter ASMR series drops Episode 2: Ollivander's shop at night — Nthorpe1996 · 2026-11-20
- One reference image, no LoRA: keeping a character consistent across 8 scenes with SnapRef — PoleRacers · 2026-11-20