[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83833-en":3,"doc-seo-83833-105":28,"detail-sidebar-cat-0-en-105":80},{"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":4,"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":11},83833,2336464648322,"Aria","https://ap-avatar.wpscdn.com/avatar/2200025388227c56fec?_k=1778556882303663488",8,"Research & Report","StructuredEdit: Constraint-Aware Graphic Design Editing via Differentiable Parameter Propagation","Graphic design editing requires precise manipulation of typography, layout, and visual hierarchy under strict design constraints. StructuredEdit addresses limitations of pixel-based vision-language models that only achieve 52% constraint satisfaction on structured design edits. The method reframes editing as parameter manipulation and introduces Differentiable Parameter Propagation (DPP) to backpropagate constraint violations through a lightweight differentiable rasterizer. A hybrid candidate-and-filter pipeline validates 125k edit triplets, improving constraint satisfaction to 89% with stronger font accuracy and lower user editing time.","StructuredEdit: Constraint-Aware Graphic Design Editing via Differentiable Parameter Propagation  \nVeeramanohar Avudaiappan  \nDepartment of Electrical and Electronics Engineering, Amrita School of Engineering, Coimbatore, Amrita Vishwa Vidyapeetham India  \n[cb.en.u4eee19145@cb.students.amrita.edu](cb.en.u4eee19145@cb.students.amrita.edu)  \nRitwik Murali∗ Department of Computer Science and Engineering, Amrita School of Computing, Coimbatore, Amrita Vishwa Vidyapeetham  \nIndia  \n[m_ritwik@cb.amrita.edu](m_ritwik@cb.amrita.edu)  \narXiv :2607 .046 12v 1 [ cs .GR] 6 Jul 2026  \nAbstract  \nGraphic design editing requires precise manipulation of typography, layout, and visual hierarchy under strict design constraints. Following the introduction of large language models, organizations have increasingly promoted vision–language models to enhance productivity. However, current models operate on pixels and achieve only 52% constraint satisfaction on structured designedits, thereby limiting their reliability for professional workflows. We present StructuredEdit, a pipeline that reframes design editing as parameter manipulation rather than pixel generation. Our core technical contribution is Differentiable Parameter Propagation (DPP), a training method that embeds hard design constraints into vision-language model fine-tuning by backpropagating pixel-level constraint violations through a lightweight differentiable rasterizer. A hybrid candidate-and-filter pipeline produces 125k validated edit triplets. The resulting system reaches 89% constraint satisfaction versus 52% for GPT-4V, 0.82 matched-element IoU, and 76% top-1 font accuracy over the 100 most-frequent design typefaces. In a user study (N=35), editing time drops 33% and correction iterations drop 44% relative to a GPT-4V baseline.  \nCCS Concepts  \n• Computing methodologies → Image manipulation; Machine learning.  \nKeywords  \nGraphic Design Editing, Differentiable Parameter Propagation, Constrained Generation, Hybrid Data Generation, Parameter-Precise Editing  \nACM Reference Format:  \nVeeramanohar Avudaiappan and Ritwik Murali. 2026. StructuredEdit: Constraint-Aware Graphic Design Editing via Differentiable Parameter Propagation. In . ACM, New York, NY, USA, 3 pages. [https://doi.org/XXXXXXX](https://doi.org/XXXXXXX). XXXXXXX  \n∗ Corresponding Author  \nPermission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission [and/or a fee. Request permissions from permissions@acm.org](and/or a fee. Request permissions from permissions@acm.org).  \nConference’17, Washington, DC, USA  \n© 2026 Copyright held by the owner/author(s) . Publication rights licensed to ACM. ACM ISBN 978-x-xxxx-xxxx-x/YYYY/MM  \n[https://doi.org/XXXXXXX.XXXXXXX](https://doi.org/XXXXXXX.XXXXXXX)  \nFigure 1: (a) A LoRA-adapted Qwen2-VL-7B predicts a parameter patch 􀀿 from a typed layer JSON and instruction, trained via cross-entropy and pixel-space constraint losses backpropagated through a differentiable rasterizer. (b) At inference, a raster poster is decomposed into structured layers, edited by the DPP-trained model, validated, and rendered.  \n1 Introduction  \nGraphic designers spend significant time on routine but fine-grained edits like resizing elements, adjusting colours, and repositioning content. Current AI tools fail at these precise, constraint-governed operations. In fact, frontier Vision-Language Models (VLMs) reach only 23.7% top-1 font identification across 167 families [Shahgir et al. 2025], and pixel-based editors introduce 29–68 px positioning errors th","cbCaisYVH9wAwuCY","https://ap.wps.com/l/cbCaisYVH9wAwuCY","pdf",1276630,1,3,"English","en",105,"# Introduction\n# Methodology","[{\"question\":\"What performance gains does StructuredEdit report compared with GPT-4V?\",\"answer\":\"StructuredEdit reports 89% constraint satisfaction versus 52% for GPT-4V, along with 0.82 matched-element IoU and 76% top-1 font accuracy over the 100 most frequent design typefaces. In a user study, editing time drops 33% and correction iterations drop 44% relative to the GPT-4V baseline.\"}]",1784190846,{"code":4,"msg":29,"data":30},"ok",{"site_id":24,"language":23,"slug":31,"title":13,"keywords":32,"description":14,"schema_data":33,"social_meta":75,"head_meta":77,"extra_data":79,"updated_unix":27},"structurededit-constraint-aware-graphic-design-editing-via-differentiable-parameter-propagation","",{"@graph":34,"@context":74},[35,51,65],{"@type":36,"itemListElement":37},"BreadcrumbList",[38,42,46,48],{"item":39,"name":40,"@type":41,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":43,"name":44,"@type":41,"position":45},"https://docshare.wps.com/document/","Document",2,{"item":47,"name":12,"@type":41,"position":21},"https://docshare.wps.com/document/research-report/",{"item":49,"name":13,"@type":41,"position":50},"https://docshare.wps.com/document/structurededit-constraint-aware-graphic-design-editing-via-differentiable-parameter-propagation/83833/",4,{"url":49,"name":13,"@type":52,"author":53,"headline":13,"publisher":55,"fileFormat":58,"inLanguage":23,"description":14,"dateModified":59,"datePublished":59,"encodingFormat":58,"isAccessibleForFree":60,"interactionStatistic":61},"DigitalDocument",{"name":9,"@type":54},"Person",{"url":39,"name":56,"@type":57},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":62,"interactionType":63,"userInteractionCount":4},"InteractionCounter",{"@type":64},"ViewAction",{"@type":66,"mainEntity":67},"FAQPage",[68],{"name":69,"@type":70,"acceptedAnswer":71},"What performance gains does StructuredEdit report compared with GPT-4V?","Question",{"text":72,"@type":73},"StructuredEdit reports 89% constraint satisfaction versus 52% for GPT-4V, along with 0.82 matched-element IoU and 76% top-1 font accuracy over the 100 most frequent design typefaces. In a user study, editing time drops 33% and correction iterations drop 44% relative to the GPT-4V baseline.","Answer","https://schema.org",{"og:url":49,"og:type":76,"og:title":13,"og:site_name":56,"og:description":14},"article",{"robots":78,"canonical":49},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":81},[82,86,90,94,99,104,109,112,117,120,124],{"id":20,"doc_module":4,"doc_module_name":44,"category_name":83,"show_sort_weight":84,"slug":85},"Story & Novel",90,"story-novel",{"id":45,"doc_module":4,"doc_module_name":44,"category_name":87,"show_sort_weight":88,"slug":89},"Literature",80,"literature",{"id":50,"doc_module":4,"doc_module_name":44,"category_name":91,"show_sort_weight":92,"slug":93},"Exam",70,"exam",{"id":95,"doc_module":4,"doc_module_name":44,"category_name":96,"show_sort_weight":97,"slug":98},5,"Comic",60,"comic",{"id":100,"doc_module":4,"doc_module_name":44,"category_name":101,"show_sort_weight":102,"slug":103},6,"Technology",50,"technology",{"id":105,"doc_module":4,"doc_module_name":44,"category_name":106,"show_sort_weight":107,"slug":108},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":44,"category_name":12,"show_sort_weight":110,"slug":111},30,"research-report",{"id":113,"doc_module":4,"doc_module_name":44,"category_name":114,"show_sort_weight":115,"slug":116},9,"Religion & Spirituality",20,"religion-spirituality",{"id":115,"doc_module":4,"doc_module_name":44,"category_name":118,"show_sort_weight":115,"slug":119},"World Cup","world-cup",{"id":121,"doc_module":4,"doc_module_name":44,"category_name":122,"show_sort_weight":121,"slug":123},10,"Lifestyle","lifestyle",{"id":125,"doc_module":4,"doc_module_name":44,"category_name":126,"show_sort_weight":95,"slug":127},19,"General","general"]