TILT: Improving Compositional Generation in Diffusion Models with a Model-Intrinsic Reward

Hugging Face Daily Papers Papers

Summary

TILT is a training-free framework that improves compositional generation in diffusion models by using a model-intrinsic reward to align sampling trajectories at test time, enhancing fidelity to complex prompts without 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.
Original Article
View Cached Full Text

Cached at: 07/28/26, 06:26 PM

Paper page - TILT: Improving Compositional Generation in Diffusion Models with a Model-Intrinsic Reward

Source: https://huggingface.co/papers/2607.21606

Abstract

Recentadvancesinpowerfultext-to-imagegenerationmodelshavemadeitincreasinglyimportanttodeveloptest-timemethodsthatmodifythesamplingtrajectorytoproduceimagesmorefaithfultocomplexcompositionalprompts.WepresentTILT,atraining-freeframeworkforcompositionaltext-to-imagegenerationviatest-timerewardalignment.Weinterpretcompositionalfailuresasoverlapmodesbetweenjointandsingle-conceptdistributions,anddefinearewardthatfavorssampleswhereallconceptsarejointlypresent.Thisrewardisintrinsictothebasemodelanddoesnotrequireanyexternalsupervisionorrewardmodels.ThisyieldsaKL-constrainedobjectivewithaclosed-formtiltedtargetdistributionandprincipledguidingstepsfordiffusionsampling.Theinteractionofconceptdistributionstogetherwiththeaboverewardnaturallyleadstotwodifferentguidancestrategieswhileahybridapproachthatbalancestheirrespectivebenefitsproducesstrongerperformance.ExperimentsonpromptsfromT2ICompBenchshowthatourmethodimprovescompositionalalignmentwhilepreservingimagequalitycomparedtopreviousbaselines.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2607\.21606

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.21606 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.21606 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.21606 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Reinforcing Few-step Generators via Reward-Tilted Distribution Matching

Hugging Face Daily Papers

RTDMD is a two-stage framework combining distribution matching distillation with reward-guided reinforcement learning to improve few-step image generation alignment with human preferences. It achieves state-of-the-art results on multiple models with only 4 inference steps.