Current approaches to 3D generation generally lack explicit local control: geometric adherence is governed by a single global control strength, and appearance cannot be specified region by region. A new research paper introduces SpaceFlow, a training-free pipeline designed to address both limitations.
SpaceFlow generates 3D content from text descriptions combined with a collection of geometric primitives. Each primitive acts as a proxy for an object part and is assigned its own local control level, letting users decide whether a given region should strictly follow the input shape or be open to generative completion.
During structure generation, the method enforces these spatial constraints within the generative flow process. For appearance synthesis, the generated structure is segmented and matched to the primitives, with each generated part conditioned only on its assigned text or image cue — a design intended to limit cross-part leakage.
According to the paper, regional geometry metrics show that SpaceFlow preserves specified geometry in high-control regions while allowing plausible shape variation in low-control areas. A user study indicates the balance between geometric fidelity and generative freedom remains competitive in overall quality. On fixed geometry, the authors report that text-conditioned routing achieves state-of-the-art prompt faithfulness and color/material accuracy, and qualitative results demonstrate localized routing of image cues.
The project page is available at SpaceFlow3D.github.io.