[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-328043-105":59,"doc-detail-328043-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","canvasagent-enabling-complex-image-creation-and-editing-via-visual-tool-orchestration","CanvasAgent - Enabling Complex Image Creation and Editing via Visual Tool Orchestration","","Complex image creation and editing often require coordinated operations beyond a single generation or editing model. A user request may involve synthesizing images, localizing and segmenting objects, editing selected regions, compositing intermediate assets, performing OCR, applying geometric transformations, and enhancing the final output. This shifts multimodal agents toward manipulation-centered visual creation with tools that actively transform visual states. This work proposes CanvasCraft and CanvasAgent, enabling multi-tool orchestration via multi-turn interaction and learning executable image-creation trajectories.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/canvasagent-enabling-complex-image-creation-and-editing-via-visual-tool-orchestration/328043/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/canvasagent-enabling-complex-image-creation-and-editing-via-visual-tool-orchestration/328043.png","ImageObject",300,407,{"name":92,"@type":93},"Eden","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-12","2026-09-21",true,{"@type":102,"interactionType":103,"userInteractionCount":44},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"What makes complex image creation and editing different from single-step editing?","Question",{"text":112,"@type":113},"Such workflows are long-horizon, visually grounded (the agent inspects intermediate outputs), and stateful, requiring correct management of multiple images, masks, crops, and extracted assets across steps.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"What are CanvasCraft-SFT and CanvasCraft-RL?",{"text":117,"@type":113},"CanvasCraft-SFT provides fully annotated executable trajectories with user instructions and step-level details. CanvasCraft-RL offers task specifications with expected tool sets, allowing reinforcement learning to explore tool ordering, parameters, recovery, and stopping.",{"name":119,"@type":110,"acceptedAnswer":120},"How does CanvasAgent learn to orchestrate visual tools during execution?",{"text":121,"@type":113},"It is first trained with SFT to learn valid executable reasoning-action trajectories, then optimized with GRPO using a hybrid reward over outcome and process signals. During rollout, it inspects intermediate results, tracks visual assets, and adapts tool decisions to the evolving visual state.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},328043,1791213959,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":44,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":139,"language":140,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":67,"update_tm":144,"read_time":145,"preview_page_count":4},1374391974468,"https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0","arXiv :2607 .05465v 1 [ cs .CV] 6 Jul 2026  \nCanvasAgent: Enabling Complex Image Creation and Editing via Visual Tool Orchestration  \nHairui Zhu1 Yiying Yang1 Tengjin Weng1 Ziyu Lu1  \nXiao Yao1 Xiaoyang Ye1 Lin Ma Wenhao Jiang1 ∗  \n1 Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ)  \nShenzhen, China  \nAbstract  \nComplex image creation and editing often require more than a single generation or editing model. A user request may involve synthesizing images, localizing objects, segmenting regions, editing selected content, compositing intermediate assets, reading text, and enhancing the final result. Such tasks shift multimodal agents from perception-augmented reasoning to manipulation-centered visual creation, where tools must actively transform visual states rather than merely inspect them. However, existing multimodal tool-use agents are mostly optimized for perception, search, or domain-specific editing, and lack large-scale supervision for executable image-creation trajectories. In this paper, we introduce CanvasCraft, a large-scale multimodal tool-use dataset for complex image creation and editing, and CanvasAgent, a tool-augmented multimodal agent that learns to orchestrate heterogeneous visual tools through multi-turn interaction. CanvasCraft contains 140K fully annotated executable trajectories and 10K RL task specifications. CanvasAgent is first trained with SFT to learn executable reasoning-action trajectories, and is then optimized with GRPO using a hybrid reward that combines outcomeand process-level signals. During rollout, CanvasAgent inspects intermediate results, tracks visual assets, and adapts tool decisions to the evolving visual state.  \nExperiments evaluate both final image quality and trajectory behavior, demonstrating the effectiveness of CanvasAgent and the proposed dataset for complex multi-tool image creation workflows.  \n1 Introduction  \nImage creation and editing have advanced rapidly with diffusion models, instruction-guided editors, and multimodal large language models (MLLMs) [13, 1, 14] . Yet many practical requests still exceed a single model call. A user may ask an assistant to generate a scene, locate and segment an object, replace only that region, add text or another object, crop the result, and enhance its resolution. Such requests require generation, localization, segmentation, editing, compositing, OCR, geometric transformation, and enhancement to be coordinated within one coherent workflow.  \nThese workflows differ from standard image generation or single-step editing in three key ways. First, they are long-horizon: later operations depend on visual artifacts produced by earlier tools. Second, they are visually grounded: after each tool call, the agent must inspect the intermediate output instead of assuming success. Third, they are stateful: multiple images, masks, crops, extracted objects, and edited variants may coexist, and the agent must select the correct asset for each subsequent operation.  \n∗Corresponding author.  \nCode and datasets are available at: [https://github.com/GML-FMGroup/CanvasAgent](https://github.com/GML-FMGroup/CanvasAgent)[ ](https://github.com/GML-FMGroup/CanvasAgent)[https://huggingface.co/datasets/GML-FMGroup/CanvasCraftSFT](https://huggingface.co/datasets/GML-FMGroup/CanvasCraftSFT)  \n[https://huggingface.co/datasets/GML-FMGroup/CanvasCraftRL](https://huggingface.co/datasets/GML-FMGroup/CanvasCraftRL).  \nThese properties make complex image creation and editing a trajectory-learning problem rather thana simple prompt-to-image or instruction-to-image task.  \nExisting multimodal tool-use research addresses parts of this problem but not the full setting. Early tool-augmented systems connect language models with visual foundation models or external experts for multi-step reasoning and editing [26, 29, 18, 6, 22] . Recent agentic MLLMs improve active visual perception, search, and executable reasoning [36, 7, 33, 24, 35], but their tasks mainly","cbCail1SDaBpAHPo","https://ap.wps.com/l/cbCail1SDaBpAHPo","pdf",2473292,18,"English","# Introduction\n## Problem setting: long-horizon, visually grounded, stateful workflows\n## Proposed solution: CanvasCraft dataset and CanvasAgent model\n## Training approach and execution protocol\n# Experiments and contributions","[{\"question\":\"What makes complex image creation and editing different from single-step editing?\",\"answer\":\"Such workflows are long-horizon, visually grounded (the agent inspects intermediate outputs), and stateful, requiring correct management of multiple images, masks, crops, and extracted assets across steps.\"},{\"question\":\"What are CanvasCraft-SFT and CanvasCraft-RL?\",\"answer\":\"CanvasCraft-SFT provides fully annotated executable trajectories with user instructions and step-level details. CanvasCraft-RL offers task specifications with expected tool sets, allowing reinforcement learning to explore tool ordering, parameters, recovery, and stopping.\"},{\"question\":\"How does CanvasAgent learn to orchestrate visual tools during execution?\",\"answer\":\"It is first trained with SFT to learn valid executable reasoning-action trajectories, then optimized with GRPO using a hybrid reward over outcome and process signals. During rollout, it inspects intermediate results, tracks visual assets, and adapts tool decisions to the evolving visual state.\"}]","CanvasAgent - Enabling Complex Image Creation and Editing via Visual Tool Orchestration | PDF",1789978571,45]