SpaceFlow: Locally Controllable 3D Generation
SpaceFlow adds local geometric and appearance control to training-free text-to-3D generation.
Researchers present SpaceFlow, a training-free pipeline for locally controllable 3D generation from text and geometric primitives. Each primitive is an object-part proxy with a local control level, so high-control regions follow the input shape while low-control regions allow generative completion. Appearance is synthesized per segmented part to limit cross-part leakage, and the authors report state-of-the-art prompt faithfulness and color and material accuracy on fixed geometry.
- Training-free 3D generation with per-primitive local control levels
- High-control regions preserve input geometry; low-control regions vary
- Part-wise text or image cues limit cross-part appearance leakage
- User study finds competitive fidelity versus generative freedom
From the original article
1 min read at arxiv.orgCurrent 3D generation methods lack explicit local control: geometric adherence is often defined by a global control strength, and appearance cannot be specified locally. We present SpaceFlow, a training-free pipeline for locally controllable 3D generation from text descriptions and a collection of geometric primitives.
By Neil De La Fuente · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.