[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-1-en-105":3,"doc-seo-238441-105":53,"doc-detail-238441-en":126},{"code":4,"msg":5,"data":6},0,"success",[7,14,19,24,29,34,39,44,49],{"id":8,"doc_module":9,"doc_module_name":10,"category_name":11,"show_sort_weight":12,"slug":13},11,1,"Template","Presentations",90,"presentations",{"id":15,"doc_module":9,"doc_module_name":10,"category_name":16,"show_sort_weight":17,"slug":18},12,"Resumes",80,"resumes",{"id":20,"doc_module":9,"doc_module_name":10,"category_name":21,"show_sort_weight":22,"slug":23},14,"Invoices",70,"invoices",{"id":25,"doc_module":9,"doc_module_name":10,"category_name":26,"show_sort_weight":27,"slug":28},15,"Posters",60,"posters",{"id":30,"doc_module":9,"doc_module_name":10,"category_name":31,"show_sort_weight":32,"slug":33},16,"Social Media",50,"social-media",{"id":35,"doc_module":9,"doc_module_name":10,"category_name":36,"show_sort_weight":37,"slug":38},17,"Forms",40,"forms",{"id":40,"doc_module":9,"doc_module_name":10,"category_name":41,"show_sort_weight":42,"slug":43},18,"Letters",30,"letters",{"id":45,"doc_module":9,"doc_module_name":10,"category_name":46,"show_sort_weight":47,"slug":48},21,"Paper Templates",5,"papers-templates",{"id":50,"doc_module":9,"doc_module_name":10,"category_name":51,"show_sort_weight":4,"slug":52},158,"General","general-158",{"code":4,"msg":54,"data":55},"ok",{"site_id":56,"language":57,"slug":58,"title":59,"keywords":60,"description":61,"schema_data":62,"social_meta":119,"head_meta":121,"extra_data":123,"updated_unix":125},105,"en","figmatrace-capturing-creative-nuances-in-human-figma-design-workflows","FIGMATRACE - Capturing Creative Nuances in Human Figma Design Workflows","","Vision Language Models improve in objective, verifiable domains yet lag on subjective, creative design tasks. A key cause is the lack of high-quality human workflow data that reflects diverse preferences and decisions underlying expert design. FIGMATRACE defines a curated taxonomy of design skills, expands it into 126 long-horizon subjective tasks, and provides 200+ hours of video converted into 3469 trajectories via a phase-based method. Training four models with FIGMATRACE yields gains on out-of-distribution agentic GUI environments and outperforms length-based conversion in ablations.",{"@graph":63,"@context":118},[64,80,101],{"@type":65,"itemListElement":66},"BreadcrumbList",[67,71,74,77],{"item":68,"name":69,"@type":70,"position":9},"https://docshare.wps.com","Home","ListItem",{"item":72,"name":10,"@type":70,"position":73},"https://docshare.wps.com/template/",2,{"item":75,"name":51,"@type":70,"position":76},"https://docshare.wps.com/template/general/",3,{"item":78,"name":59,"@type":70,"position":79},"https://docshare.wps.com/template/figmatrace-capturing-creative-nuances-in-human-figma-design-workflows/238441/",4,{"url":78,"name":59,"@type":81,"image":82,"author":87,"headline":59,"publisher":90,"fileFormat":93,"inLanguage":57,"description":61,"dateModified":94,"datePublished":95,"encodingFormat":93,"isAccessibleForFree":96,"interactionStatistic":97},"DigitalDocument",{"url":83,"@type":84,"width":85,"height":86},"https://docshare.wps.com/thumbnails/figmatrace-capturing-creative-nuances-in-human-figma-design-workflows/238441.png","ImageObject",442,249,{"name":88,"@type":89},"Sophia Brooks","Person",{"url":68,"name":91,"@type":92},"DocShare","Organization","application/pdf","2026-09-25","2026-09-11",true,{"@type":98,"interactionType":99,"userInteractionCount":73},"InteractionCounter",{"@type":100},"ViewAction",{"@type":102,"mainEntity":103},"FAQPage",[104,110,114],{"name":105,"@type":106,"acceptedAnswer":107},"Why do vision language models underperform on creative design tasks?","Question",{"text":108,"@type":109},"They struggle to model subjective nuances because high-quality human workflow data capturing real preferences and design decisions is largely unavailable.","Answer",{"name":111,"@type":106,"acceptedAnswer":112},"What does FIGMATRACE provide?",{"text":113,"@type":109},"FIGMATRACE offers an expert-curated taxonomy of design skills and a dataset built from 200+ hours of human-captured Figma workflow video converted into 3469 design trajectories.",{"name":115,"@type":106,"acceptedAnswer":116},"How does FIGMATRACE’s video-to-trajectory conversion work, and why is it important?",{"text":117,"@type":109},"It uses a novel design phase-based conversion method, and ablation results attribute performance improvements to this phase-based approach rather than length-based conversion.","https://schema.org",{"og:url":78,"og:type":120,"og:title":59,"og:site_name":91,"og:description":61},"article",{"robots":122,"canonical":78},"index,follow",{"doc_id":124,"site_id":56},238441,1790359926,{"code":4,"msg":5,"data":127},{"doc_id":124,"user_id":128,"nickname":88,"user_avatar":129,"doc_module":9,"category_id":50,"category_name":51,"doc_title":59,"doc_description":61,"doc_content":130,"file_id":131,"file_url":132,"file_type":133,"file_size":134,"view_count":73,"is_deleted":4,"is_public":9,"is_downloadable":9,"audit_status":9,"page_count":30,"language":135,"language_code":57,"site_id":56,"html_lang":57,"table_of_contents":136,"faqs":137,"seo_title":138,"seo_description":61,"update_tm":139,"read_time":140},962084925636,"https://ap-avatar.wpscdn.com/davatar_994ba38a5ba835b3df7d355c54d3ed8d","arXiv : 2608 .2 1460v 1 [cs .CV] 20 Aug 2026  \nFIGMATRACE: CAPTURING CREATIVE NUANCES INHUMAN FIGMA DESIGN WORKFLOWS  \nDarshan DeshpandeYoshinari Fujinuma, Martyna Markiewicz, Devanshu Bansal  \nShivani Jain, Nicholas Saban, Chirag Maheshwari, Anand Kannappan  Patronus AI  \n{darshan, yoshinari .fujinuma, martyna, dev, shivani, nicksaban, chirag.maheshwari, [anand](anand}@patronus.ai)[}](anand}@patronus.ai)[@patronus.ai](anand}@patronus.ai)  \nABSTRACT  \nVision Language Models have recently shown improvements in several objective and verifiable domains such as object detection but continue to underperform on subjective and creative design tasks. A major contributor to this performance gap is the lack of high quality human workflow data that captures a diverse set of preferences and decisions that make human experts good at design tasks. In this work, we first define a unique, expert curated taxonomy of design skills and best practices which we further expand into a set of 126 open ended, subjective, long horizon tasks. Built on top of this and expert solutions, our dataset FIGMATRACE contains over 200 hours of human captured video data converted into 3469 design trajectories using a novel design phase-based method. We use our dataset to train four models and show that training on FIGMATRACE leads to a performance improvement comparable to frontier closed models such as CLAUDE-OPUS-5 and GPT-5.6-SOL on four out of distribution agentic GUI environments. We further perform a useful ablation to attribute these performance improvements to a design phase-based video to trajectory conversion which outperforms prior length-based conversion approaches. Finally, we perform a qualitative analysis on the best performing QWEN 3.8-27B outputs to better correlate performance improvements to FIGMATRACE’s trends. We open source our dataset and the best QWEN 3.8- 27B model for the community 1.  \n1 INTRODUCTION  \nVision Language Models (VLMs) are popularly used for several verifiable tasks such as document understanding (Ding et al., 2026; Wang et al., 2025a), robotics (Sapkota et al., 2025; Zhang et al., 2025) and computer use (Tang et al., 2025; Xie et al., 2025) . On subjective and non-verifiable tasks, VLMs have struggled to capture human nuances such as understanding emotions (Bhattacharyya & Wang, 2025), humor and understanding of figurative meaning (Zhou et al., 2026; Ryan et al., 2025) and design taste (An et al., 2026) . Recent works have attempted to address this issue through human preference alignment (Peng et al., 2025; Liao et al., 2025) and reinforcement learning based objectives (Li et al., 2025b; Wu et al., 2026), however, these techniques rely heavily on the availability of data that surfaces such preferences. This is worsened by the unavailability of high quality design datasets used to train tasteful and nuanced design agents.  \nTo address this lack of data, prior works such as Gui et al. (2026) explored using existing Figmadesigns and working backwards to create automated annotation processes for data but the validation of such work is difficult and ambiguous due to the lack of exhaustive quality guidelines. Exploring verifiability, Russo et al. (2025) use existing HTML versions of pages to convert them to Figma compatible JSONs that a model can be trained to generate (human grounding for LLMs is difficult + validation is hard) . Kanapathipillai & Priyankara (2026) explore a parallel direction of automated creation of individual Figma component JSONs for the sake of reusability and scalability. Making  \n∗ Correspondence: [darshan@patronus.ai](darshan@patronus.ai)  \n1 [https://huggingface.co/datasets/PatronusAI/figmatrace](https://huggingface.co/datasets/PatronusAI/figmatrace)[ ](https://huggingface.co/datasets/PatronusAI/figmatrace)[https://huggingface.co/PatronusAI/Qwen3.8-27B-Figmatrace-SFT](https://huggingface.co/PatronusAI/Qwen3.8-27B-Figmatrace-SFT)  \n(a) Set hexcodekeyboard type(\"00DC82\")  \n(b) Create stylemouse   click(1","cbCaijVG3mdC5l1n","https://ap.wps.com/l/cbCaijVG3mdC5l1n","pdf",1455473,"English","# Abstract\n# Introduction\n## Research Gap in Vision Language Models\n## Prior Data-Creation and Evaluation Approaches\n## Proposed FIGMATRACE Dataset and Method","[{\"question\":\"Why do vision language models underperform on creative design tasks?\",\"answer\":\"They struggle to model subjective nuances because high-quality human workflow data capturing real preferences and design decisions is largely unavailable.\"},{\"question\":\"What does FIGMATRACE provide?\",\"answer\":\"FIGMATRACE offers an expert-curated taxonomy of design skills and a dataset built from 200+ hours of human-captured Figma workflow video converted into 3469 design trajectories.\"},{\"question\":\"How does FIGMATRACE’s video-to-trajectory conversion work, and why is it important?\",\"answer\":\"It uses a novel design phase-based conversion method, and ablation results attribute performance improvements to this phase-based approach rather than length-based conversion.\"}]","FIGMATRACE - Capturing Creative Nuances in Human Figma Design Workflows | PDF",1789135554,6]