UniForge3D open-sourced: joint 3D reconstruction and generation from sparse multi-view images
victormustar · x · 2026-10-26
UniForge3D (xibushu/UniForge3D, CC-BY-NC-4.0) is now on Hugging Face with a demo Space: it generates complete 3D assets from sparse, unposed multi-view images.
Key ideas:
- Jointly models reconstruction and generation in a shared anchor-guided space, balancing observed consistency with generative completion
- Uses structure-guided view selection to preserve details actually seen in the inputs
An Image-to-3D model; no Inference Provider deployed yet, but usable via its Space.
More from Multimodal
- Claude rebuilds a Gaussian-splatted garden scene in Blender with Geometry Nodes — CSProfKGD · 2026-10-26
- AI film festival veteran publishes 19 suggestions for organizers and submitters — lmoroney · 2026-10-26
- Three prompts for Claude grid-effect music videos: stretching 15s footage to 40s — alexcovo_eth · 2026-10-26
- Midjourney character sheet demo shows impressive multi-angle consistency — Salmaaboukarr · 2026-10-26
- Imagination Studio: rebuilt local open-source platform for video, audio, music and image generation — Scobleizer · 2026-10-26
- Beeble SwitchX 2 swaps video backgrounds and relights footage so well this creator is ditching the podcast studio — venturetwins · 2026-10-26