[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83103-en":3,"doc-seo-83103-105":30,"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":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},83103,1099514067415,"Rowan","https://ap-avatar.wpscdn.com/avatar/100002539d78ffe74a7?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779092875211072502",8,"Research & Report","PIPBench A Profile-Inclusive Framework for Personalized Image Generation Evaluation","Personalized image generation aligns outputs with a user’s implicit aesthetic preferences derived from a few historically preferred images plus a short prompt. The work introduces PIPBench, the first profile-inclusive benchmark designed to evaluate this setting using real user profiles paired with preferred images. A novel data construction pipeline uses psychological and demographic profiling dimensions to support both real-user data collection and scalable agent-based synthetic data generation. Experiments across representative personalized text-to-image methods identify key limitations and surface new challenges and opportunities for personalized generation.","arXiv :2607 .06440v 1 [ cs .CV] 7 Jul 2026  \nCollege of AI, Tsinghua University  \nJuly 2, 2026  \nPIPBench: A Profile-Inclusive Framework for Personalized Image Generation Evaluation  \nYuhang Wu 1 Shuxiang Zhang2 WEE HIAN CHING 1 Chi Zhang 1,3 Miao Liu 1†  \n1 College of AI, Tsinghua University  \n2 Sun Yat-sen University  \n3 Shanghai Qi Zhi Institute  \nProject Page Code Dataset  \nAbstract. Recent text-to-image models such as DALL·E-3 excel at following diverse prompts yet remain blind to individual aesthetic preferences. We study personalized image generation, where models must align outputs with a user’s implicit visual preferences based on a few historically preferred images and a short prompt. To this end, we introduce PIPBench, the first profile-inclusive benchmark for evaluating personalized image generation. We further propose a novel data construction pipeline that leverages psychological and demographic profiling dimensions for both real-user data collection and scalable agent-based data generation. Using PIPBench, we conduct a thorough evaluation of representative line of methods. Our experiments reveal key limitations in existing methods, suggesting new challenges and opportunities for personalized text-to-image synthesis.  \nKeywords: Personalized generation, Image generation, Benchmark  \nUser Implicit Preference  \nUser-preferred Images  \na penguin standing alone on the ice  \na kitten playing with a ball of yarn  \nperson with windblown hair on a coastal path by the sea  \na green cup of tea and a pastry on a windowsill  \nUser1  \nminimalist cool vibes  \ncel-shaded anime  \nquiet  \nocean lover  \nUser3  \nprecision hardware  \ncreativity  \nengineering tech-savvy  \nFigure 1 . Problem setup of personalized image generation and examples from our PIPBench. Our benchmark incorporates both user profiles and preferred images, enabling a comprehensive analysis of implicit and explicit visual aesthetic preferences.  \n1 Introduction  \nRecent text-to-image generation models, such as DALL·E 3 [3], have attracted widespread attention. However, these systems [3 , 10 , 35] remain largely task-driven: they readily follow explicit user instructions, yet adopt a generic generative model without accounting for implicit individual preferences.  \nConsider the examples in Fig. 1, a typical user prompt conveys only the core concept. Existing systems first expand such prompts using large language models (LLMs), and then generate images based on enriched descriptions. This two-stage procedure introduces an enormous search space due to the stochasticity of both language enrichment and image synthesis. However, each user may resonate with only a small portion of this exploration space, resulting in a high interaction barrier for non-expert users, since personal preferences are often implicit and difficult to express through plain language.  \nAn alternative paradigm for personalized text-to-image generation is illustrated in Fig. 1. Specifically, the generation model leverages historical user preference data to produce results that are better aligned with the user’s aesthetic intent. This setting relates to prior work on style transfer and image editing, which align visual synthesis with visual or textual guidance. However, these methods do not consider user aesthetic preference modeling.  \nNotably, only a handful recent studies [16, 30 , 38] have examined image generation conditioned on user preference data, largely due to the absence of a systematic benchmark with high-quality user data. Existing datasets [6, 7 , 13] that include user interaction histories are largely domain-specific, focusing on applications such as sticker or poster preferences. In contrast, for general personalized image generation, individual implicit visual preferences are inherently difficult to articulate and quantify. For example, a user may consistently favor images depicting open natural landscapes over enclosed urban scenes, without being consciously aware of this tendency.","cbCail47LYF0OITx","https://ap.wps.com/l/cbCail47LYF0OITx","pdf",25992479,2,1,39,"English","en",105,"# Introduction\n## Motivation and problem setup\n## Related paradigms and limitations\n## Contributions\n# Related Work\n## Personalized modeling and generation","[{\"question\":\"What problem does PIPBench address in personalized image generation?\",\"answer\":\"It evaluates how well text-to-image models can align generated images with a user’s implicit visual preferences from limited historical preferred images and a short prompt.\"},{\"question\":\"How does PIPBench incorporate user information into the benchmark?\",\"answer\":\"The benchmark pairs real user profiles with their corresponding preferred images, enabling analysis of both implicit and explicit visual aesthetic preferences.\"},{\"question\":\"What is the purpose of the proposed agent-based data construction pipeline?\",\"answer\":\"It leverages psychological and demographic profiling dimensions to support diverse benchmark data distribution, including scalable synthetic data generation via agents.\"}]",1784185273,98,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"pipbench-a-profile-inclusive-framework-for-personalized-image-generation-evaluation","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/pipbench-a-profile-inclusive-framework-for-personalized-image-generation-evaluation/83103/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-21","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 PIPBench address in personalized image generation?","Question",{"text":75,"@type":76},"It evaluates how well text-to-image models can align generated images with a user’s implicit visual preferences from limited historical preferred images and a short prompt.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does PIPBench incorporate user information into the benchmark?",{"text":80,"@type":76},"The benchmark pairs real user profiles with their corresponding preferred images, enabling analysis of both implicit and explicit visual aesthetic preferences.",{"name":82,"@type":73,"acceptedAnswer":83},"What is the purpose of the proposed agent-based data construction pipeline?",{"text":84,"@type":76},"It leverages psychological and demographic profiling dimensions to support diverse benchmark data distribution, including scalable synthetic data generation via agents.","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":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]