T2I models still struggle with complex prompts—like getting attribute binding, spatial relations, or counting wrong.<br>To bridge this gap, we proposed BiDPO. We curate a high-quality preference dataset (BiComp), extend Diffusion DPO to jointly optimize image &amp; text preferences, and apply region-level guidance for fine-grained alignment. It delivers a huge boost in compositional fidelity.<br>🔥 We’ve already open-sourced the dataset and model weights right here on Hugging Face. Check them out!<br>🎉 Super excited to share that this work is accepted to CVPR 2026!</p>\n","updatedAt":"2026-06-02T02:36:14.334Z","author":{"_id":"655b2dfe9d249b4ab38ca25f","avatarUrl":"/avatars/86250cb9ebbd1d2330607ff41be61d31.svg","fullname":"Zhuohan Liu","name":"anzeameol","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.885520875453949},"editors":["anzeameol"],"editorAvatarUrls":["/avatars/86250cb9ebbd1d2330607ff41be61d31.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.28615","authors":[{"_id":"6a19107a56b4bb14ec65d015","user":{"_id":"655b2dfe9d249b4ab38ca25f","avatarUrl":"/avatars/86250cb9ebbd1d2330607ff41be61d31.svg","isPro":false,"fullname":"Zhuohan Liu","user":"anzeameol","type":"user","name":"anzeameol"},"name":"Zhuohan Liu","status":"admin_assigned","statusLastChangedAt":"2026-06-01T17:43:26.507Z","hidden":false},{"_id":"6a19107a56b4bb14ec65d016","name":"Wujian Peng","hidden":false},{"_id":"6a19107a56b4bb14ec65d017","name":"Yitong Chen","hidden":false},{"_id":"6a19107a56b4bb14ec65d018","name":"Zuxuan Wu","hidden":false}],"publishedAt":"2026-05-27T00:00:00.000Z","submittedOnDailyAt":"2026-06-02T00:00:00.000Z","title":"Compositional Text-to-Image Generation Via Region-aware Bimodal Direct Preference Optimization","submittedOnDailyBy":{"_id":"655b2dfe9d249b4ab38ca25f","avatarUrl":"/avatars/86250cb9ebbd1d2330607ff41be61d31.svg","isPro":false,"fullname":"Zhuohan Liu","user":"anzeameol","type":"user","name":"anzeameol"},"summary":"Despite the rapid progress of text-to-image (T2I) models, generating images that accurately reflect complex compositional prompts (covering attribute bindings, object relationships, counting) still remains challenging. To address this, we propose BiDPO, a framework to enhance T2I model's capability of compositional text-to-image generation. We begin by introducing an carefully designed pipeline to construct a large-scale preference dataset, BiComp, with strictly quality control. Then, we extend Diffusion DPO to jointly optimize image and text preferences, which is shown to greatly effective in improving the models to follow complex text prompt in generation. To further enhance the models for fine-grained alignment, we employ a region-level guidance method to focus on regions relevant to compositional concepts. Experimental results demonstrate that our BiDPO substantially improves compositional fidelity, consistently outperforming prior methods across multiple benchmarks. Our approach highlights the potential of preference-based fine-tuning for complex text-to-image tasks, offering a flexible and scalable alternative to existing techniques.","upvotes":2,"discussionId":"6a19107a56b4bb14ec65d019","githubRepo":"https://github.com/anzeameol/BiDPO","githubRepoAddedBy":"user","ai_summary":"BiDPO enhances text-to-image models for complex compositional prompts through preference-based fine-tuning and region-level guidance.","ai_keywords":["text-to-image models","compositional prompts","preference-based fine-tuning","Diffusion DPO","preference dataset","BiComp","region-level guidance","compositional fidelity"],"githubStars":2},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"699eb03bdd2252e76a2bddbb","avatarUrl":"/avatars/e48e17bbcebda53f347849af91af0234.svg","isPro":false,"fullname":"Jaxon Williams","user":"jaxonwilliams20","type":"user"},{"_id":"69bcb8ec9044732b67dac308","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/blygEHY5DTVU8UsuEBKMN.png","isPro":false,"fullname":"Григорьев Даниил","user":"chloeallen","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.28615.md"}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2605.28615 in a Space README.md to link it from this page.

Add this paper to a collection to link it from this page.