[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82218-en":3,"doc-seo-82218-105":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},82218,1374391974468,"Eden","https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0",6,"Technology","Exploring the Potential of Program Flowcharts on Code Generation Using Multimodal LLMs","Recent advances in large language models (LLMs) have enabled multimodal systems that process images and other inputs alongside text. Prior work suggests that adding visual signals can strengthen automatic code generation, yet the specific effect of software flowcharts remains insufficiently studied. This work generates flowcharts from AtCoder example solutions and provides them with problem statements to GPT-4o. Results show up to 10% performance gains, with finer-grained flowcharts correlating with higher accuracy and cost comparisons favoring one-shot improvements.","arXiv :2607 .09 146v 1 [ cs . SE] 10 Jul 2026  \nExploring the Potential of Program Flowcharts on Code Generation Using Multimodal LLMs  \nYuki Toi 1[0009−0000−4567−1607], Tao Xiao 1 􀀌 [0000−0003−4070−585X], Kazushi Tomoto 1[0009−0009−7455−6304], and Masanari Kondo 1[0000−0002−6317−7001], and Yasutaka Kamei 1[0000−0002−7058−1045]  \nKyushu University, Fukuoka, Japan, {toi,[tomoto}@posl.ait.kyushu-u.ac.jp](tomoto}@posl.ait.kyushu-u.ac.jp) ,{xiao,kondo,[kamei}@ait.kyushu-u.ac.jp](kamei}@ait.kyushu-u.ac.jp)  \nAbstract. In recent years, Large Language Models (LLMs) have made significant strides, leading to the emergence of multimodal LLMs capable of processing diverse inputs such as images and audio. Previous research indicates that the supply of multimodal LLMs with combined textual and visual information improves the automatic code generation capabilities. In software development, diagrams such as flowcharts are widely employed to facilitate tasks like code comprehension.  \nWhile existing studies investigated the impact of visual inputs on LLMs and the usage of software diagrams, the potential influence of providing flowcharts on multimodal LLM performance remains underexplored. In this study, we generated flowcharts from example solution code for AtCoder problems and provided these visual aids alongside problem statements to GPT-4o for code generation.  \nOur findings demonstrate that integrating flowcharts with problem statements yields performance improvements of up to 10% . Furthermore, when employing abstracted flowcharts, we observed a trend indicating that increasing levels of flowchart detail correlate with enhanced performance. Additionally, we compared the effectiveness of flowchart provision to Few-Shot Learning approaches.  \nThe findings suggest that one-shot learning provides sustainable improvements, whereas two-shot learning results in only minor improvements. Our work highlights the importance of software diagrams in supporting multimodal LLM-driven code generation.  \nKeywords: Large Language Model · Automatic Code Generation · Flowchart.  \n1 Introduction  \nCode generation, which involves the automated production of source code from natural language descriptions, represents a cornerstone in the field of automated software engineering. The early approaches to this task were based on heuristic rules [1, 13, 31] and expert systems [3, 7, 11] . However, these foundational techniques were typically rigid and difficult to scale. The advent of Transformer-based Large Language Models (LLMs) [28] has shifted the paradigm of code generation from the reliance on heuristic rule-based systems to more powerful, scalable approaches that learn patterns from vast code repositories [12] . This domain garnered substantial interest from academic and  \n2 Toi et al.  \nindustrial experts, as evidenced by the development and widespread adoption of tools such as GitHub Copilot 1 [21, 35] and ChatGPT2 [4, 18] .  \nThe practice of software development involves not only textual artifacts, such as source code and documentation, but also essential visual elements such as diagrams and user interfaces. Therefore, LLMs have been developing, with multimodal LLMs capable of recognizing images and speech in addition to text. GPT-4o,3 developed by OpenAI, is one such multimodal LLM; it can be used interactively with ChatGPT or integrated into a system using the API. These multimodal LLMs have been used for web interface design [26], diagramming, and visual testing [32] .  \nThe performance of LLMs on code generation tasks can be enhanced through the integration of multimodal support. To this end, Li et al. [16] developed the MMCode benchmark to assess the algorithmic problem-solving capabilities of models in visually rich contexts, using GPT-4V4 and Gemini Pro Vision.5 They observed improved results when a visual context of the problem was supplied alongside its textual description. Nevertheless, the study concludes that successful problem-solving hinges ","cbCaiuEZgjhc7aVe","https://ap.wps.com/l/cbCaiuEZgjhc7aVe","pdf",427473,1,21,"English","en",105,"# Introduction\n## Background on code generation and multimodal LLMs\n## Research gap: flowcharts for visual program logic\n# Method\n## Dataset reuse from AtCoder problems\n## Flowchart generation and prompt construction\n## Visual abstraction and multi-granularity evaluation\n# Findings\n## Performance improvements with flowchart support\n## Effect of flowchart detail level\n## Comparison with Few-Shot Learning","[{\"question\":\"How are flowcharts used in the code generation prompts for GPT-4o?\",\"answer\":\"Flowcharts are generated from AtCoder example solution code and provided alongside the problem statements. GPT-4o is evaluated on prompts containing only text versus prompts augmented with flowcharts.\"},{\"question\":\"What performance improvement does flowchart integration achieve?\",\"answer\":\"Integrating flowcharts with problem statements yields performance improvements of up to around 10%, and the benefit increases in line with problem difficulty.\"},{\"question\":\"How does flowchart abstraction level affect results compared with Few-Shot Learning?\",\"answer\":\"Abstracted flowcharts show a trend where more detailed flowcharts correlate with better performance. One-shot learning provides more sustainable improvements, while two-shot learning leads to only minor gains.\"}]",1784178917,53,{"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},"exploring-the-potential-of-program-flowcharts-on-code-generation-using-multimodal-llms","",{"@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/technology/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/exploring-the-potential-of-program-flowcharts-on-code-generation-using-multimodal-llms/82218/",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},"How are flowcharts used in the code generation prompts for GPT-4o?","Question",{"text":75,"@type":76},"Flowcharts are generated from AtCoder example solution code and provided alongside the problem statements. GPT-4o is evaluated on prompts containing only text versus prompts augmented with flowcharts.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What performance improvement does flowchart integration achieve?",{"text":80,"@type":76},"Integrating flowcharts with problem statements yields performance improvements of up to around 10%, and the benefit increases in line with problem difficulty.",{"name":82,"@type":73,"acceptedAnswer":83},"How does flowchart abstraction level affect results compared with Few-Shot Learning?",{"text":84,"@type":76},"Abstracted flowcharts show a trend where more detailed flowcharts correlate with better performance. One-shot learning provides more sustainable improvements, while two-shot learning leads to only minor gains.","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},[93,97,101,105,110,113,118,123,128,131,135],{"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":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":111,"slug":112},50,"technology",{"id":114,"doc_module":4,"doc_module_name":45,"category_name":115,"show_sort_weight":116,"slug":117},7,"Healthcare",40,"healthcare",{"id":119,"doc_module":4,"doc_module_name":45,"category_name":120,"show_sort_weight":121,"slug":122},8,"Research & Report",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"]