Text-to-image models rely on text prompts as their primary interface to human intent. Prompts are encoded by a text encoder into embeddings that condition the image generation process. Beyond individual token meanings, text embeddings encode contextual information across the full prompt, such as compositionality and attribute binding. However, whether image models actually exploit this richer information remains underexplored. Here, we address the question: Which aspects of text representation are essential for image generation? We show that text-to-image diffusion transformer-based models commonly rely only on two relatively straightforward aspects of text representations: (i) the merging of adjacent tokens into a word representation, for words spanning multiple tokens, and (ii) word order, which is imprinted by the positional embedding of the text-encoder. To show this, we construct a new text embedding that encodes only individual word meanings and order but lacks any contextual information about the full prompt. We find that this bag of position-tagged words representation is sufficient to successfully guide image generation, achieving visual quality and text fidelity that are on par with full text embedding-guided generation. This demonstrates that, contrary to common belief, text-to-image models often do not use the rich information encoded in the text embedding beyond individual word meanings and word order. Instead, the decoding of complex linguistic structures is performed by the image model itself.</p>\n","updatedAt":"2026-06-09T07:19:13.005Z","author":{"_id":"6632355191e7cf5661c11bd5","avatarUrl":"/avatars/c2bc424d1810cd6e37d87afded93de0a.svg","fullname":"Noa","name":"cohennoa","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9005946516990662},"editors":["cohennoa"],"editorAvatarUrls":["/avatars/c2bc424d1810cd6e37d87afded93de0a.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.03715","authors":[{"_id":"6a27b7a36dde1c5ef75bd1bc","name":"Nurit Spingarn","hidden":false},{"_id":"6a27b7a36dde1c5ef75bd1bd","name":"Noa Cohen","hidden":false},{"_id":"6a27b7a36dde1c5ef75bd1be","name":"Tamar Rott Shaham","hidden":false},{"_id":"6a27b7a36dde1c5ef75bd1bf","name":"Tomer Michaeli","hidden":false}],"publishedAt":"2026-06-02T00:00:00.000Z","submittedOnDailyAt":"2026-06-09T00:00:00.000Z","title":"Text-to-Image Models Need Less from Text Encoders Than You Think","submittedOnDailyBy":{"_id":"6632355191e7cf5661c11bd5","avatarUrl":"/avatars/c2bc424d1810cd6e37d87afded93de0a.svg","isPro":false,"fullname":"Noa","user":"cohennoa","type":"user","name":"cohennoa"},"summary":"Text-to-image models rely on text prompts as their primary interface to human intent. Prompts are encoded by a text encoder into embeddings that condition the image generation process. Beyond individual token meanings, text embeddings encode contextual information across the full prompt, such as compositionality and attribute binding. However, whether image models actually exploit this richer information remains underexplored. Here, we address the question: Which aspects of text representation are essential for image generation? We show that text-to-image diffusion transformer-based models commonly rely only on two relatively straightforward aspects of text representations: (i) the merging of adjacent tokens into a word representation, for words spanning multiple tokens, and (ii) word order, which is imprinted by the positional embedding of the text-encoder. To show this, we construct a new text embedding that encodes only individual word meanings and order but lacks any contextual information about the full prompt. We find that this bag of position-tagged words representation is sufficient to successfully guide image generation, achieving visual quality and text fidelity that are on par with full text embedding-guided generation. This demonstrates that, contrary to common belief, text-to-image models often do not use the rich information encoded in the text embedding beyond individual word meanings and word order. Instead, the decoding of complex linguistic structures is performed by the image model itself. Project webpage: https://nsping13.github.io/contextless-TTI/","upvotes":5,"discussionId":"6a27b7a36dde1c5ef75bd1c0","projectPage":"https://nsping13.github.io/contextless-TTI/","ai_summary":"Text-to-image models primarily utilize basic text representation aspects like word merging and order rather than complex contextual information encoded in full text embeddings.","ai_keywords":["text-to-image models","text encoder","text embeddings","diffusion transformer-based models","positional embedding","bag of position-tagged words","visual quality","text fidelity"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"63511e3aa8822aadf5720ff9","avatarUrl":"/avatars/5798938f1c90d642b2d492b67c77aa4c.svg","isPro":false,"fullname":"Matan Kleiner","user":"matankleiner","type":"user"},{"_id":"6632355191e7cf5661c11bd5","avatarUrl":"/avatars/c2bc424d1810cd6e37d87afded93de0a.svg","isPro":false,"fullname":"Noa","user":"cohennoa","type":"user"},{"_id":"668d51b0a825eb9e6ed44a71","avatarUrl":"/avatars/641e94491a9f8639a2475d7d70cfb47a.svg","isPro":false,"fullname":"Yakov Fridman","user":"YakovFridman","type":"user"},{"_id":"63f6454a9cbd673030295df4","avatarUrl":"/avatars/90c0e4a1c5dffdabbf2492853741d0d9.svg","isPro":false,"fullname":"Hila Manor","user":"hilamanor","type":"user"},{"_id":"63f6497eb29015adc33c1da7","avatarUrl":"/avatars/9119dffcec465b42c490d0e893699a60.svg","isPro":false,"fullname":"Vladimir Kulikov","user":"fallenshock","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.03715.md"}">
Text-to-Image Models Need Less from Text Encoders Than You Think
Published on Jun 2
· Submitted by Noa on Jun 9 Abstract
Text-to-image models primarily utilize basic text representation aspects like word merging and order rather than complex contextual information encoded in full text embeddings.
Text-to-image models rely on text prompts as their primary interface to human intent. Prompts are encoded by a text encoder into embeddings that condition the image generation process. Beyond individual token meanings, text embeddings encode contextual information across the full prompt, such as compositionality and attribute binding. However, whether image models actually exploit this richer information remains underexplored. Here, we address the question: Which aspects of text representation are essential for image generation? We show that text-to-image diffusion transformer-based models commonly rely only on two relatively straightforward aspects of text representations: (i) the merging of adjacent tokens into a word representation, for words spanning multiple tokens, and (ii) word order, which is imprinted by the positional embedding of the text-encoder. To show this, we construct a new text embedding that encodes only individual word meanings and order but lacks any contextual information about the full prompt. We find that this bag of position-tagged words representation is sufficient to successfully guide image generation, achieving visual quality and text fidelity that are on par with full text embedding-guided generation. This demonstrates that, contrary to common belief, text-to-image models often do not use the rich information encoded in the text embedding beyond individual word meanings and word order. Instead, the decoding of complex linguistic structures is performed by the image model itself. Project webpage: https://nsping13.github.io/contextless-TTI/
Community
Text-to-image models rely on text prompts as their primary interface to human intent. Prompts are encoded by a text encoder into embeddings that condition the image generation process. Beyond individual token meanings, text embeddings encode contextual information across the full prompt, such as compositionality and attribute binding. However, whether image models actually exploit this richer information remains underexplored. Here, we address the question: Which aspects of text representation are essential for image generation? We show that text-to-image diffusion transformer-based models commonly rely only on two relatively straightforward aspects of text representations: (i) the merging of adjacent tokens into a word representation, for words spanning multiple tokens, and (ii) word order, which is imprinted by the positional embedding of the text-encoder. To show this, we construct a new text embedding that encodes only individual word meanings and order but lacks any contextual information about the full prompt. We find that this bag of position-tagged words representation is sufficient to successfully guide image generation, achieving visual quality and text fidelity that are on par with full text embedding-guided generation. This demonstrates that, contrary to common belief, text-to-image models often do not use the rich information encoded in the text embedding beyond individual word meanings and word order. Instead, the decoding of complex linguistic structures is performed by the image model itself.
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/2606.03715 in a model README.md to link it from this page.
Cite arxiv.org/abs/2606.03715 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2606.03715 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.