SpatialCritic improves rewards for better image generation with curriculum learning
Getting Space Right: Agentic Rewards for Image Generation
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Generating images from text can be tricky when the image needs to match spatial instructions, like placing objects in the right places. The authors found that earlier methods struggled because they assumed rewards and feedback were accurate and progressive, which isn’t always true in real-world settings. They designed a new system called SpatialCritic that uses special visual tools to check spatial details accurately and learns through a step-by-step curriculum to improve results over time. This makes image generators better at following spatial cues in their outputs.
What this means in practice
- •For graphic designers: Generate images from spatially detailed text prompts with better adherence to object placement instructions.
- •For game developers: Create visual scenes from descriptive text while ensuring objects appear in the correct spatial arrangements as intended.
Authors
Xiaofeng Tan, Jiaxin Lin, Haobo Yuan, Tiancheng Shen, Zihan Su, Rui Chen, Hongsong Wang, Ming-Hsuan Yang
Abstract
Reinforcement learning (RL) is promising for spatial text-to-image generation, but existing work largely focuses on policy optimization, assuming accurate rewards and effective online rollouts. We observe in-domain gains with out-of-domain degradation, exposing two limitations: unreliable spatial rewards in open-world settings and rollouts lacking basic-to-complex progression. We propose a curriculum-guided RL framework. First, we introduce SpatialCritic, an agentic reward model equipped with a spatial perception harness that combines reusable verification guidance with specialized vision tools. We then construct spatial preference datasets through targeted counterfactual perturbations and perform agent-harness alignment to improve tool selection and evidence integration efficiently. Second, curriculum learning organizes shared foundational capabilities and their compositional dependencies in a capability graph for mastery-guided rollouts, periodic review, and targeted remediation. On SpaRW-Eval, SpatialCritic achieves 95.1% verification accuracy, outperforming 320B GLM-5.3-Flash with a 40x smaller 8B backbone and raising average baseline scores from 78.4 to 83.9 on SD3.5-Medium and 83.2 to 86.5 on BAGEL. Project page: https://xiaofeng-tan.github.io/projects/spatialcritic/