Exploring Spatial Intelligence from a Generative Perspective

TL;DR

This paper introduces GSI-Bench, a new benchmark to evaluate and improve generative spatial intelligence in multimodal models through spatially grounded image editing tasks.

Contribution

It presents GSI-Bench, including real and synthetic datasets, and demonstrates that fine-tuning models on this benchmark enhances both generative and spatial understanding capabilities.

Findings

Fine-tuning on GSI-Syn improves model performance on synthetic tasks.

Fine-tuning on GSI-Syn enhances real-world spatial reasoning.

Generative training can strengthen spatial intelligence in multimodal models.

Abstract

Spatial intelligence is essential for multimodal large language models, yet current benchmarks largely assess it only from an understanding perspective. We ask whether modern generative or unified multimodal models also possess generative spatial intelligence (GSI), the ability to respect and manipulate 3D spatial constraints during image generation, and whether such capability can be measured or improved. We introduce GSI-Bench, the first benchmark designed to quantify GSI through spatially grounded image editing. It consists of two complementary components: GSI-Real, a high-quality real-world dataset built via a 3D-prior-guided generation and filtering pipeline, and GSI-Syn, a large-scale synthetic benchmark with controllable spatial operations and fully automated labeling. Together with a unified evaluation protocol, GSI-Bench enables scalable, model-agnostic assessment of spatial compliance and editing fidelity. Experiments show that fine-tuning unified multimodal models on GSI-Syn yields substantial gains on both synthetic and real tasks and, strikingly, also improves downstream spatial understanding. This provides the first clear evidence that generative training can tangibly strengthen spatial reasoning, establishing a new pathway for advancing spatial intelligence in multimodal models.