[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84184-en":3,"doc-seo-84184-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},84184,1374391974468,"Eden","https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0",8,"Research & Report","Tree of Thoughts Reasoning for Text to Image In-Context Learning","Text-to-image in-context learning (T2I-ICL) requires a model to infer latent compositional structure from few-shot demonstrations to generate a query image. Existing multimodal LLMs show limited compositional reasoning and sensitivity to prompt construction. This work proposes a Tree-of-Thoughts (ToT) multi-stage reasoning and selection framework that generates, evaluates, and chooses among candidate hypotheses before prompt creation. Evaluations on the CoBSAT benchmark show more consistent, semantically aligned image generation than baseline and Chain-of-Thought prompting, without extra training.","Tree-of-Thoughts Reasoning for Text-to-Image In-Context Learning  \nStepanida Alekseeva 1 , Jenifer Kalafatovich 1 and Seong-Whan Lee 1 ,∗  \narXiv :2607 .07 1 17v 1 [ cs .CV] 8 Jul 2026  \nAbstract—In text-to-image in-context learning (T2I-ICL), a model has to infer a latent compositional pattern from fewshot demonstrations for generating a query image. Recent studies show that state-of-the-art multimodal large language models struggle with this setting, particularly due to limited compositional reasoning and sensitivity to prompt construction. In this work, we propose a Tree-of-Thoughts (ToT) reasoning framework for T2I-ICL that introduces a multi-stage reasoning and selection layer that generates, evaluates, and selects among multiple candidate hypotheses before constructing the final prompt for image synthesis. By exploring alternative reasoning branches and selecting a coherent interpretation, the proposed approach mitigates prompt ambiguity and compositional errors. We implement the proposed approach in a complete ToT-T2IICL inference pipeline and evaluate it on the CoBSAT benchmark. Both qualitative and quantitative results show that structured multi-branch reasoning leads to more consistent and semantically aligned image generation compared to baseline and Chain-of-Thought prompting strategies, without any additional training or fine-tuning. Note: The code is publicly available at [https://github.com/Pandastep/ToT-T2I-ICL](https://github.com/Pandastep/ToT-T2I-ICL).  \nI. INTRODUCTION  \nText-to-image (T2I) generation aims to synthesize images from natural language descriptions, enabling intuitive human-AI interaction and controllable visual content creation. Recent advances in T2I generation have enabled highquality image synthesis from natural language descriptions [1], [2] . Despite strong visual fidelity, modern models still struggle with compositional generalization, particularly incorrectly binding objects and attributes [3] .  \nIn-context learning (ICL) enables models to adapt to new tasks and patterns at inference time without parameter updates, making it a flexible and efficient alternative to taskspecific training. While ICL has been widely adopted in multimodal large language models (MLLMs) for tasks like captioning and visual question answering [4], [5], its application in T2I remains underexplored. Recent benchmarks such as CoBSAT reveal that MLLMs face notable difficulties in T2IICL, indicating limited capacity for structured compositional reasoning [6] .  \nIn this paper, we address this limitation by introducing a Tree-of-Thoughts (ToT) reasoning framework for T2I-ICL. Instead of relying on a single reasoning trajectory as in  \nThis work was supported by the Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No. RS-2019-II190079, Artificial Intelligence Graduate School Program, Korea University), and by the Information Technology Research Center (ITRC) support program (No. IITP-2026-RS- 2024-00436857) .  \n1Department of Artificial Intelligence, Korea University, Seoul, Korea. e-mail: [alexeeva.stepanida@gmail.com](alexeeva.stepanida@gmail.com); [jenifer@korea.ac.kr](jenifer@korea.ac.kr); [sw.lee@korea.ac.kr](sw.lee@korea.ac.kr)  \n∗Corresponding author.  \nprevious methods, the proposed approach explores multiple candidate interpretations, evaluates them with respect to the in-context examples, and selects a coherent hypothesis for image generation. This structured reasoning process improves robustness to ambiguity and reduces compositional errors. Our method operates entirely at inference time and does not require model fine-tuning. Specifically, the model first analyzes the input demonstrations, then constructs a prompt through ToT reasoning, and finally generates animage based on the resulting prompt. This separation allows us to isolate the effect of reasoning on generation quality.  \nThe main contributions of this work are as follow","cbCaihmYuTN7TJHB","https://ap.wps.com/l/cbCaihmYuTN7TJHB","pdf",414252,1,6,"English","en",105,"# Introduction\n## Text-to-image and controllability\n## In-context learning for T2I\n# Related Works\n## Text-to-Image generation and controllability\n## In-Context Learning for Text-to-Image Generation","[{\"question\":\"What problem does T2I-ICL face in current multimodal LLMs?\",\"answer\":\"Models struggle to infer compositional generalization from few-shot demonstrations, often incorrectly binding objects and attributes. Performance is also sensitive to how prompts are constructed.\"},{\"question\":\"How does the proposed Tree-of-Thoughts (ToT) framework work for T2I-ICL?\",\"answer\":\"It explores multiple reasoning branches to produce candidate interpretations, evaluates them using the in-context examples, then selects a coherent hypothesis. Only after selection does it construct the final prompt for image synthesis.\"},{\"question\":\"What improvements are reported on the CoBSAT benchmark?\",\"answer\":\"Structured multi-branch reasoning yields more consistent and semantically aligned image generation than baseline and Chain-of-Thought strategies. The approach achieves this using an inference-time pipeline without additional training or fine-tuning.\"}]",1784193727,15,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"tree-of-thoughts-reasoning-for-text-to-image-in-context-learning","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/tree-of-thoughts-reasoning-for-text-to-image-in-context-learning/84184/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does T2I-ICL face in current multimodal LLMs?","Question",{"text":75,"@type":76},"Models struggle to infer compositional generalization from few-shot demonstrations, often incorrectly binding objects and attributes. Performance is also sensitive to how prompts are constructed.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the proposed Tree-of-Thoughts (ToT) framework work for T2I-ICL?",{"text":80,"@type":76},"It explores multiple reasoning branches to produce candidate interpretations, evaluates them using the in-context examples, then selects a coherent hypothesis. Only after selection does it construct the final prompt for image synthesis.",{"name":82,"@type":73,"acceptedAnswer":83},"What improvements are reported on the CoBSAT benchmark?",{"text":84,"@type":76},"Structured multi-branch reasoning yields more consistent and semantically aligned image generation than baseline and Chain-of-Thought strategies. The approach achieves this using an inference-time pipeline without additional training or fine-tuning.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,114,119,122,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":21,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]