[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85423-en":3,"doc-seo-85423-105":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},85423,1099513958762,"Logic","https://ap-avatar.wpscdn.com/avatar/1000023916a998db790?x-image-process=image/resize,m_fixed,w_180,h_180&k=1782109480056885918",8,"Research & Report","TIIF-Bench: How Does Your T2I Model Follow Your Instructions?","Text-to-Image (T2I) models have improved their ability to interpret and follow user instructions, yet existing evaluation benchmarks limit prompt diversity, complexity, and metric granularity, hindering precise assessment of instruction-image alignment. TIIF-Bench (Text-to-Image Instruction Following Benchmark) provides 5,000 prompts across multiple dimensions and three difficulty levels. Each prompt is offered in short and long forms to test robustness to length, and a Global Normalized Edit Distance metric evaluates text rendering. Designer-level prompts, aspect-ratio-diverse references, and automated binary evaluators from large vision-language models enable scalable, interpretable, reproducible evaluation, revealing fine-grained strengths and weaknesses.","TIIF-Bench: How Does Your T2I Model Follow Your Instructions?  \nXinyu Wei1 ,3 , Jinrui Zhang1 ,3 , Zeqing Wang1 ,3 , Hongyang Wei2 ,3 Zhen Guo1 ,3 , Bairui Li1 ,3 , and Lei Zhang1 ,3  \n1 The Hong Kong Polytechnic University  \n2 Tsinghua University  \n3 OPPO Research Institute  \nAbstract. The rapid advancements of Text-to-Image (T2I) models have ushered in a new phase of AI-generated content, marked by their growing ability to interpret and follow user instructions. However, existing T2I model evaluation benchmarks fall short in limited prompt diversity and complexity, as well as coarse evaluation metrics, making it difficult to evaluate the fine-grained alignment performance between textual instructions and generated images. In this paper, we present TIIF-Bench (Textto-Image Instruction Following Benchmark), aiming to systematically assess T2I models’ ability in interpreting and following intricate textual instructions. TIIF-Bench comprises 5,000 prompts organized along multiple dimensions and categorized into three levels of difficulty and complexity. To rigorously evaluate robustness to prompt length, each prompt is provided in both short and long versions with identical core semantics. We further propose a novel Global Normalized Edit Distance (GNED) metric for text rendering and provide aspect-ratio-diverse reference images for each prompt to assess style control. In addition, we collect 100 high-quality designer-level prompts covering diverse scenarios for comprehensive evaluation. To enable scalable and fine-grained evaluation, we explore the best paradigm for leveraging the world knowledge encoded in large Vision-Language Models (VLMs) as automated binary evaluators. Through extensive ablations, we develop a fully reproducible evaluator that provides interpretable reasoning and reliable verification, enabling our benchmark to discern subtle variations in T2I model outputs. Through comprehensive benchmarking of mainstream T2I modelson TIIF-Bench, we analyze the strengths and weaknesses of current T2I systems and reveal the limitations of existing evaluation benchmarks.  \nKeywords: T2I Generation · Benchmarking · Reward Signal  \nIntroduction  \nText-to-Image (T2I) generation has emerged as a cornerstone of multimodal AI, enabling the translation of abstract textual concepts into detailed visual  \n2 Authors Suppressed Due to Excessive Length  \nFig. 1: Prompt-length sensitivity example. Short prompt: “The birds are more numerous than the fish.” Long prompt: “The birds, with their feathers catching the gentle light of dawn, vastly outnumber their aquatic counterparts, the fish, which glide silently beneath the rippling surface of the water, their sleek forms moving like shadows in the depths below.” PixArt-Sigma and SD 3.5 show clear sensitivity to prompt length, producing different results for semantically equivalent prompts. In contrast, DALL·E 3 and GPTImage-1 remain robust, maintaining consistent instruction-following performance.  \ncontent, advancing applications from digital art to scientific visualization. Recent T2I models can be categorized into 3 main paradigms. Diffusion-based methods—exemplified by Stable Diffusion [17, 49], PixArt [7, 10], FLUX [30], SANA [9, 70], Qwen-Image [66], and others [4, 14, 28, 37, 40, 56, 81]—leverage U-Net or Diffusion-Transformer backbones to iteratively denoise Gaussian noise into photorealistic images, achieving strong visual fidelity and diversity. Autoregressive (AR) approaches, such as LlamaGen [53], Infinity [22], VAR [55], and other influential open-source models, treat images as token sequences and synthesize them through next-token prediction or scale-progressive generation. Unified-model (UM) approaches, including the Show-o series [71, 72], Janus [67] and JanusPro [12], the Lumina-mGPT series [42, 73], Bagel [16] and the EMU series [15], unify generation and understanding within a single transformer backbone, aiming to mutually enhance multimodal reasoning and image synthesi","cbCaihdbmeINo6G9","https://ap.wps.com/l/cbCaihdbmeINo6G9","pdf",31022979,1,35,"English","en",105,"# Introduction\n## Prompt-length sensitivity and robustness\n## T2I model paradigms\n## Existing evaluation approaches","[{\"question\":\"What problem does TIIF-Bench address in evaluating T2I models?\",\"answer\":\"It addresses the lack of prompt diversity/complexity and coarse evaluation metrics in existing benchmarks, which makes fine-grained alignment between textual instructions and generated images hard to measure.\"},{\"question\":\"How is TIIF-Bench designed to test robustness to prompt length?\",\"answer\":\"Each of the 5,000 prompts is provided in both short and long versions while keeping the core semantics identical, allowing direct comparison of instruction-following behavior under different lengths.\"},{\"question\":\"What metric and evaluation mechanism does TIIF-Bench introduce?\",\"answer\":\"TIIF-Bench proposes a Global Normalized Edit Distance (GNED) metric for text rendering and uses large vision-language models as automated binary evaluators to provide interpretable reasoning and reliable verification.\"}]",1784203359,88,{"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},"tiif-bench-how-does-your-t2i-model-follow-your-instructions","",{"@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/tiif-bench-how-does-your-t2i-model-follow-your-instructions/85423/",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 TIIF-Bench address in evaluating T2I models?","Question",{"text":75,"@type":76},"It addresses the lack of prompt diversity/complexity and coarse evaluation metrics in existing benchmarks, which makes fine-grained alignment between textual instructions and generated images hard to measure.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How is TIIF-Bench designed to test robustness to prompt length?",{"text":80,"@type":76},"Each of the 5,000 prompts is provided in both short and long versions while keeping the core semantics identical, allowing direct comparison of instruction-following behavior under different lengths.",{"name":82,"@type":73,"acceptedAnswer":83},"What metric and evaluation mechanism does TIIF-Bench introduce?",{"text":84,"@type":76},"TIIF-Bench proposes a Global Normalized Edit Distance (GNED) metric for text rendering and uses large vision-language models as automated binary evaluators to provide interpretable reasoning and reliable verification.","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},[93,97,101,105,110,115,120,123,128,131,135],{"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":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]