AI TrackerFrontier desk · AEST
Back to news
normalaitech

SpaceFlow: Locally Controllable 3D Generation

arXiv cs.AI · 9 Oct 2026, 03:42 AEST

SpaceFlow introduces a training-free pipeline for locally controllable 3D generation from text descriptions and geometric primitives. It allows users to specify control levels for different regions, enhancing both geometric fidelity and generative freedom.

Key facts

Read aloud

SpaceFlow introduces a training-free pipeline for locally controllable 3D generation from text descriptions and geometric primitives. It allows users to specify control levels for different regions, enhancing both geometric fidelity and generative freedom. Current 3D generation methods lack explicit local control: geometric adherence is often defined by a global control strength, and appearance cannot be specified locally. Each primitive serves as a proxy for an object part and is assigned a local control level, enabling users to specify whether regions should strictly follow the input shape or allow generative completion. Regional geometry metrics demonstrate that SpaceFlow preserves the specified geometry in high-control regions and enables plausible shape variation in low-control areas. When evaluating appearance on fixed geometry, text-conditioned routing achieves state-of-the-art prompt faithfulness and color/material accuracy.

Sources in this cluster