Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces TILT, a novel framework that enhances compositional text-to-image generation in diffusion models by employing a test-time reward alignment strategy. By interpreting compositional failures as overlaps between joint and single-concept distributions, TILT defines an intrinsic reward that encourages the presence of all concepts in generated images without requiring external supervision. Experimental results demonstrate that TILT significantly improves compositional alignment and maintains high image quality compared to existing methods, showcasing its effectiveness on prompts from T2ICompBench.
TILT reveals that intrinsic rewards can dramatically enhance compositional fidelity in text-to-image generation without any external supervision.
Recent advances in powerful text-to-image generation models have made it increasingly important to develop test-time methods that modify the sampling trajectory to produce images more faithful to complex compositional prompts. We present TILT, a training-free framework for compositional text-to-image generation via test-time reward alignment. We interpret compositional failures as overlap modes between joint and single-concept distributions, and define a reward that favors samples where all concepts are jointly present. This reward is intrinsic to the base model and does not require any external supervision or reward models. This yields a KL-constrained objective with a closed-form tilted target distribution and principled guiding steps for diffusion sampling. The interaction of concept distributions together with the above reward naturally leads to two different guidance strategies while a hybrid approach that balances their respective benefits produces stronger performance. Experiments on prompts from T2ICompBench show that our method improves compositional alignment while preserving image quality compared to previous baselines.