Hugging Face Daily Papers · June 24, 2026 · 4 min read

Semantic Browsing: Controllable Diversity for Image Generation

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

Semantic Browsing introduces an agentic workflow that turns a single text prompt into a structured, browsable gallery of diverse image interpretations, where each variation reflects meaningful and controllable semantic choices rather than stochastic sampling.</p>\n","updatedAt":"2026-06-24T13:30:17.537Z","author":{"_id":"64ff145e1a7cb7c342ae0c53","avatarUrl":"/avatars/5f1b8e809cfa5b5ff2b882948a934cb7.svg","fullname":"Omer Dahary","name":"omer11a","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8433429598808289},"editors":["omer11a"],"editorAvatarUrls":["/avatars/5f1b8e809cfa5b5ff2b882948a934cb7.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.23679","authors":[{"_id":"6a3bda125240b1765ae59394","name":"Sara Dorfman","hidden":false},{"_id":"6a3bda125240b1765ae59395","user":{"_id":"67127a1602d4213c4abe7a87","avatarUrl":"/avatars/e6ae050bceb58c1b1c36c7cef6a99721.svg","isPro":false,"fullname":"Maya","user":"mayavish","type":"user","name":"mayavish"},"name":"Maya Vishnevsky","status":"claimed_verified","statusLastChangedAt":"2026-06-24T15:42:27.553Z","hidden":false},{"_id":"6a3bda125240b1765ae59396","name":"Omer Dahary","hidden":false},{"_id":"6a3bda125240b1765ae59397","name":"Or Patashnik","hidden":false},{"_id":"6a3bda125240b1765ae59398","name":"Daniel Cohen-Or","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/64ff145e1a7cb7c342ae0c53/xPaxpLc8K5z9Po1LLrYOJ.png"],"publishedAt":"2026-06-22T00:00:00.000Z","submittedOnDailyAt":"2026-06-24T00:00:00.000Z","title":"Semantic Browsing: Controllable Diversity for Image Generation","submittedOnDailyBy":{"_id":"64ff145e1a7cb7c342ae0c53","avatarUrl":"/avatars/5f1b8e809cfa5b5ff2b882948a934cb7.svg","isPro":false,"fullname":"Omer Dahary","user":"omer11a","type":"user","name":"omer11a"},"summary":"Modern text-to-image models excel in visual fidelity and prompt adherence. However, this strict adherence comes at the cost of diversity: generated samples tend to collapse into a single visual interpretation. Existing methods to improve diversity produce outputs driven by incidental variations rather than meaningful design choices. This motivates a new variant of the diversity task where structure is enforced on the generated samples. We introduce a method for controlled diversity that enables Semantic Browsing, where users can navigate structured image galleries and experience creative exploration through a systematic traversal of meaningful, interpretable axes of variation. Achieving this level of semantic control requires a deep understanding of the scene. We exploit the fact that recent text-to-image models are trained on elaborated captions, effectively decoupling semantic decision-making from pixel generation. This enables a paradigm shift: instead of relying on stochastic variation within the text-to-image model, we induce diversity directly at the text level. By leveraging rich textual representations, we allow a Vision Language Model (VLM) to operate on the full scene context. To overcome the generic outputs typical of standard VLMs, we employ an agentic workflow that explicitly enforces structured variation attuned to the original prompt. We demonstrate that our method produces diverse and navigable design spaces where every variation corresponds to a specific, user-understandable semantic decision.","upvotes":13,"discussionId":"6a3bda135240b1765ae59399","projectPage":"https://saradorfman1.github.io/SemanticBrowsing-webpage/","ai_summary":"Text-to-image models are enhanced with controlled diversity through semantic browsing capabilities that enable structured navigation of image variations based on meaningful semantic decisions.","ai_keywords":["text-to-image models","semantic browsing","controlled diversity","Vision Language Model","agentic workflow","semantic decision-making","scene context","structured variation","prompt adherence","visual fidelity"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64ff145e1a7cb7c342ae0c53","avatarUrl":"/avatars/5f1b8e809cfa5b5ff2b882948a934cb7.svg","isPro":false,"fullname":"Omer Dahary","user":"omer11a","type":"user"},{"_id":"668569d06be8cd35515b9be2","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/668569d06be8cd35515b9be2/50Nggue0FIZfCrQi_8x7P.jpeg","isPro":false,"fullname":"Sara Dorfman","user":"saradorfman","type":"user"},{"_id":"64646a88a5af935cfe741d3b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64646a88a5af935cfe741d3b/DEmHne74aRyd3XjHz82_M.png","isPro":true,"fullname":"shelly golan","user":"shellygolan","type":"user"},{"_id":"630b37dee67c604e9b79e773","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/630b37dee67c604e9b79e773/roSaYAsQ4E9rQrahYZitG.png","isPro":false,"fullname":"Sagi Polaczek","user":"SagiPolaczek","type":"user"},{"_id":"65fbf768980d4143b0d3ab52","avatarUrl":"/avatars/018a26ce944860e9ab97ca43765e7d76.svg","isPro":false,"fullname":"Daniel Garibi","user":"garibida","type":"user"},{"_id":"62853516e483e0d37b354ce1","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/62853516e483e0d37b354ce1/t5Tyd3E07w26B9Z3XpZWI.jpeg","isPro":false,"fullname":"Or Patashnik","user":"orpatashnik","type":"user"},{"_id":"67127a1602d4213c4abe7a87","avatarUrl":"/avatars/e6ae050bceb58c1b1c36c7cef6a99721.svg","isPro":false,"fullname":"Maya","user":"mayavish","type":"user"},{"_id":"60dd3fe45c42407cc9fe5953","avatarUrl":"/avatars/6d4a4faa6852824b8e8186585b76b09b.svg","isPro":false,"fullname":"Ariel Bereslavsky","user":"HyperActiveMind","type":"user"},{"_id":"636f56beaf79dd193603a243","avatarUrl":"/avatars/009d0fa45ec827592735a4ab2c927cb7.svg","isPro":false,"fullname":"Dana Cohen Bar","user":"dana-cohen","type":"user"},{"_id":"610b89ccf5098560ee5fa698","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1628156505064-610b89ccf5098560ee5fa698.jpeg","isPro":false,"fullname":"Adi Haviv","user":"adihaviv","type":"user"},{"_id":"661fecbb9c1169698537623f","avatarUrl":"/avatars/3a4453dc962558fa110fcdd31866f831.svg","isPro":false,"fullname":"Guy Levy","user":"George331","type":"user"},{"_id":"6613206695c6b73065c0bcc8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/U6igoyYyfFU4p0sAgzcrw.jpeg","isPro":false,"fullname":"alpaim","user":"alpaim","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.23679.md","query":{}}">

Papers

arxiv:2606.23679

Semantic Browsing: Controllable Diversity for Image Generation

Published on Jun 22

· Submitted by

Omer Dahary on Jun 24

Upvote

Authors:

Maya Vishnevsky ,

Abstract

Text-to-image models are enhanced with controlled diversity through semantic browsing capabilities that enable structured navigation of image variations based on meaningful semantic decisions.

Generated by Qwen/Qwen2.5-Coder-32B-Instruct

Modern text-to-image models excel in visual fidelity and prompt adherence. However, this strict adherence comes at the cost of diversity: generated samples tend to collapse into a single visual interpretation. Existing methods to improve diversity produce outputs driven by incidental variations rather than meaningful design choices. This motivates a new variant of the diversity task where structure is enforced on the generated samples. We introduce a method for controlled diversity that enables Semantic Browsing, where users can navigate structured image galleries and experience creative exploration through a systematic traversal of meaningful, interpretable axes of variation. Achieving this level of semantic control requires a deep understanding of the scene. We exploit the fact that recent text-to-image models are trained on elaborated captions, effectively decoupling semantic decision-making from pixel generation. This enables a paradigm shift: instead of relying on stochastic variation within the text-to-image model, we induce diversity directly at the text level. By leveraging rich textual representations, we allow a Vision Language Model (VLM) to operate on the full scene context. To overcome the generic outputs typical of standard VLMs, we employ an agentic workflow that explicitly enforces structured variation attuned to the original prompt. We demonstrate that our method produces diverse and navigable design spaces where every variation corresponds to a specific, user-understandable semantic decision.