SmartSpatial: Enhancing 3D Spatial Awareness in Stable Diffusion with a Novel Evaluation Framework
Mao Xun Huang, Brian J Chan, Hen-Hsen Huang · 2025
Read SmartSpatial: Enhancing 3D Spatial Awareness in Stable Diffusion with a Novel Evaluation Framework by Mao Xun Huang, Brian J Chan, Hen-Hsen Huang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-01998.txt.
Description
Text-to-image models draw pretty scenes but misplace objects in three dimensions, stacking chairs inside tables or floating cups. Mao Xun Huang, Brian Chan, and Hen-Hsen Huang attack that weakness with SmartSpatial, which feeds depth information into the generator and steers its attention so objects land where the prompt puts them.
They pair the fix with an evaluation framework that scores spatial accuracy computationally and folds in artistic judgement, since a technically correct jumble can still be bad art. Across tests the enhanced model beat existing approaches on spatial fidelity benchmarks. The authors frame the pair as tooling for AI-assisted creative work, where a designer's spatial intent deserves respect from the machine.
Catalog record
| CREATOR | Mao Xun Huang, Brian J Chan, Hen-Hsen Huang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-01998