<a href=\"https://arxiv.org/pdf/2606.24548\" rel=\"nofollow\">https://arxiv.org/pdf/2606.24548</a></p>\n","updatedAt":"2026-06-24T02:21:22.901Z","author":{"_id":"625d5b9f0bec31f086e04cd9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1650285458447-noauth.jpeg","fullname":"YuandongPu","name":"Andrew613","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":4,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.5253946185112},"editors":["Andrew613"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1650285458447-noauth.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.24548","authors":[{"_id":"6a3b3ee70a86ac3098d5d6d4","name":"Jiayi Lei","hidden":false},{"_id":"6a3b3ee70a86ac3098d5d6d5","name":"Yuandong Pu","hidden":false},{"_id":"6a3b3ee70a86ac3098d5d6d6","name":"Xingyu Han","hidden":false},{"_id":"6a3b3ee70a86ac3098d5d6d7","name":"Rongpeng Zhu","hidden":false},{"_id":"6a3b3ee70a86ac3098d5d6d8","name":"Jing Xu","hidden":false},{"_id":"6a3b3ee70a86ac3098d5d6d9","name":"Jinyao Wang","hidden":false},{"_id":"6a3b3ee70a86ac3098d5d6da","name":"Zijian Zhou","hidden":false},{"_id":"6a3b3ee70a86ac3098d5d6db","name":"Bin Fu","hidden":false},{"_id":"6a3b3ee70a86ac3098d5d6dc","name":"Yuewen Cao","hidden":false},{"_id":"6a3b3ee70a86ac3098d5d6dd","name":"Yihao Liu","hidden":false},{"_id":"6a3b3ee70a86ac3098d5d6de","name":"Yongsheng Li","hidden":false}],"publishedAt":"2026-06-23T00:00:00.000Z","submittedOnDailyAt":"2026-06-24T00:00:00.000Z","title":"Are Text-to-Image Models Inductivist Turkeys? A Counterfactual Benchmark for Causal Reasoning","submittedOnDailyBy":{"_id":"625d5b9f0bec31f086e04cd9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1650285458447-noauth.jpeg","isPro":false,"fullname":"YuandongPu","user":"Andrew613","type":"user","name":"Andrew613"},"summary":"Text-to-image (T2I) generation models have achieved remarkable progress in producing visually realistic images from natural language prompts. Yet it remains unclear whether their success reflects genuine causal understanding or sophisticated pattern matching over visual-textual correlations. Inspired by Russell's inductivist turkey, we introduce Counterfactual-World (CF-World), a counterfactual benchmark designed to investigate whether text-to-image models can generate images under rules that systematically contradict real-world priors. CF-World organizes each scenario into three progressive levels: factual generation under ordinary world knowledge, explicit counterfactual generation with direct visual instructions, and implicit counterfactual generation requiring causal deduction from altered rules. We evaluate both open-source and closed-source T2I models using a Vision Language Model (VLM)-based evaluator (CF-Eval). Furthermore, we introduce two metrics: Prior Resistance Rate (PRR), which measures a model's ability to overcome entrenched real-world priors, and Reasoning Retention Rate (RRR), which assesses whether models can maintain reasoning-dependent counterfactual generation without explicit visual cues. Experiments show that all models exhibit sharp degradation from factual to counterfactual settings. Further analyses suggest that these failures arise because current T2I models encode world knowledge and visual appearances as tightly coupled patterns. Consequently, their heavy reliance on frequent visual co-occurrences within the training data forces them to default to familiar commonsense priors when tasked with rendering counterfactual worlds.","upvotes":4,"discussionId":"6a3b3ee80a86ac3098d5d6df","projectPage":"https://jylei16.github.io/CF-World.github.io/","ai_summary":"Text-to-image models fail to generate counterfactual scenes because they rely on tightly coupled visual-textual patterns rather than causal reasoning, demonstrating limited understanding beyond pattern matching.","ai_keywords":["text-to-image generation","counterfactual benchmark","Vision Language Model","CF-Eval","Prior Resistance Rate","Reasoning Retention Rate","causal understanding","pattern matching","visual-textual correlations","world knowledge encoding"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"625d5b9f0bec31f086e04cd9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1650285458447-noauth.jpeg","isPro":false,"fullname":"YuandongPu","user":"Andrew613","type":"user"},{"_id":"662885b9b87483ae5a9ee5c9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/662885b9b87483ae5a9ee5c9/fLzCrWizo_hy7zGuAqFwk.jpeg","isPro":false,"fullname":"Songhao Han","user":"hshjerry0315","type":"user"},{"_id":"64c3c72e8f31d1e6c664b052","avatarUrl":"/avatars/af1ad5048eaa9dc417837ad02f927911.svg","isPro":false,"fullname":"jiayi lei","user":"jyjyjyjy","type":"user"},{"_id":"634ec067aae4bde2c8dfc86f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/634ec067aae4bde2c8dfc86f/OQBLKcspofUqAzmEpvH0-.png","isPro":false,"fullname":"Yamata Zen","user":"yamatazen","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.24548.md","query":{}}">
Are Text-to-Image Models Inductivist Turkeys? A Counterfactual Benchmark for Causal Reasoning
Authors: ,
,
,
,
,
,
,
,
,
,
Abstract
Text-to-image models fail to generate counterfactual scenes because they rely on tightly coupled visual-textual patterns rather than causal reasoning, demonstrating limited understanding beyond pattern matching.
Text-to-image (T2I) generation models have achieved remarkable progress in producing visually realistic images from natural language prompts. Yet it remains unclear whether their success reflects genuine causal understanding or sophisticated pattern matching over visual-textual correlations. Inspired by Russell's inductivist turkey, we introduce Counterfactual-World (CF-World), a counterfactual benchmark designed to investigate whether text-to-image models can generate images under rules that systematically contradict real-world priors. CF-World organizes each scenario into three progressive levels: factual generation under ordinary world knowledge, explicit counterfactual generation with direct visual instructions, and implicit counterfactual generation requiring causal deduction from altered rules. We evaluate both open-source and closed-source T2I models using a Vision Language Model (VLM)-based evaluator (CF-Eval). Furthermore, we introduce two metrics: Prior Resistance Rate (PRR), which measures a model's ability to overcome entrenched real-world priors, and Reasoning Retention Rate (RRR), which assesses whether models can maintain reasoning-dependent counterfactual generation without explicit visual cues. Experiments show that all models exhibit sharp degradation from factual to counterfactual settings. Further analyses suggest that these failures arise because current T2I models encode world knowledge and visual appearances as tightly coupled patterns. Consequently, their heavy reliance on frequent visual co-occurrences within the training data forces them to default to familiar commonsense priors when tasked with rendering counterfactual worlds.
Community
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/2606.24548 in a model README.md to link it from this page.
Cite arxiv.org/abs/2606.24548 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2606.24548 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.