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.</p>\n","updatedAt":"2026-07-28T17:28:17.748Z","author":{"_id":"68be23ac5271b9ac99d6f956","avatarUrl":"/avatars/26360a5957f6ffa6d18f335e18e940dc.svg","fullname":"Debottam Dutta","name":"debottamd07","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9014869332313538},"editors":["debottamd07"],"editorAvatarUrls":["/avatars/26360a5957f6ffa6d18f335e18e940dc.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2607.21606","authors":[{"_id":"6a68e5b65dad3683ba934988","name":"Debottam Dutta","hidden":false},{"_id":"6a68e5b65dad3683ba934989","name":"Jaehoon Hahm","hidden":false},{"_id":"6a68e5b65dad3683ba93498a","name":"Jianchong Chen","hidden":false},{"_id":"6a68e5b65dad3683ba93498b","name":"Romit Roy Choudhury","hidden":false}],"publishedAt":"2026-05-16T00:00:00.000Z","submittedOnDailyAt":"2026-07-28T00:00:00.000Z","title":"TILT: Improving Compositional Generation in Diffusion Models with a Model-Intrinsic Reward","submittedOnDailyBy":{"_id":"68be23ac5271b9ac99d6f956","avatarUrl":"/avatars/26360a5957f6ffa6d18f335e18e940dc.svg","isPro":false,"fullname":"Debottam Dutta","user":"debottamd07","type":"user","name":"debottamd07"},"summary":"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.","upvotes":0,"discussionId":"6a68e5b65dad3683ba93498c"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[],"acceptLanguages":["en"],"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2607/2607.21606.md","query":{}}">
TILT: Improving Compositional Generation in Diffusion Models with a Model-Intrinsic Reward
Abstract
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.
Community
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.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2607.21606 in a model README.md to link it from this page.
Cite arxiv.org/abs/2607.21606 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2607.21606 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.