UGG unifies video generation and 3D reconstruction in a single generative model
du_yilun · x · 2026-11-17
UGG unifies video generation and 3D reconstruction in one generative model.
- Jointly models video, depth and camera pose, treating a scene as one generative target
- Can infer missing parts of a scene and generate entirely new scenes
- Improves geometric predictions via inference-time search
- Accepted to NeurIPS 2026; co-first authors are Zhiyi Li and Ruojin
- Project page is live
More from Multimodal
- Kling 4.0 Image lands on fal with native 8K, 2,000-char typography, 10-image consistency — gorkem · 2026-11-17
- Mirelo launches Sound Agent: prompt-based sound design, dubbing and music swap for video — Scobleizer · 2026-11-17
- AI-Generated Short Film 'Blaskets' Delivers Moody Cinematic Sea Voyage — Papercrabstudio · 2026-11-17
- 'xdots' Free AI Video Generation Demo Shows Strikingly Realistic Everyday Scenes — Sufficient-Bug-104 · 2026-11-17
- French creator builds an AI-made GTA-style Marseille while waiting for GTA 6 — Dimillian · 2026-11-17
- Savannah nocturne: a surreal AI-generated safari scene with cinematic lighting — cosmoxtraveller · 2026-11-17