Hugging Face Daily Papers · June 2, 2026 · 4 min read

Crafter: A Multi-Agent Harness for Editable Scientific Figure Generation from Diverse Inputs

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

Crafter is a multi-agent system for generating publication-quality scientific figures across diverse types and conditions, with CraftEditor turning raster outputs into editable SVGs and CraftBench for evaluation.</p>\n","updatedAt":"2026-06-02T02:10:36.220Z","author":{"_id":"637c99bbfe115289cfedfb44","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/637c99bbfe115289cfedfb44/p4uSY0TKufJfcHpvEb_ZQ.jpeg","fullname":"ssz","name":"ssz1111","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":5,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.8724867701530457},"editors":["ssz1111"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/637c99bbfe115289cfedfb44/p4uSY0TKufJfcHpvEb_ZQ.jpeg"],"reactions":[{"reaction":"👀","users":["ssz1111"],"count":1}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.30611","authors":[{"_id":"6a1e3aa8808ddbc3c7d43bc1","name":"Haozhe Zhao","hidden":false},{"_id":"6a1e3aa8808ddbc3c7d43bc2","name":"Shuzheng Si","hidden":false},{"_id":"6a1e3aa8808ddbc3c7d43bc3","name":"Zhenhailong Wang","hidden":false},{"_id":"6a1e3aa8808ddbc3c7d43bc4","name":"Zheng Wang","hidden":false},{"_id":"6a1e3aa8808ddbc3c7d43bc5","name":"Liang Chen","hidden":false},{"_id":"6a1e3aa8808ddbc3c7d43bc6","name":"Xiaotong Li","hidden":false},{"_id":"6a1e3aa8808ddbc3c7d43bc7","name":"Zhixiang Liang","hidden":false},{"_id":"6a1e3aa8808ddbc3c7d43bc8","name":"Maosong Sun","hidden":false},{"_id":"6a1e3aa8808ddbc3c7d43bc9","name":"Minjia Zhang","hidden":false}],"publishedAt":"2026-05-28T00:00:00.000Z","submittedOnDailyAt":"2026-06-02T00:00:00.000Z","title":"Crafter: A Multi-Agent Harness for Editable Scientific Figure Generation from Diverse Inputs","submittedOnDailyBy":{"_id":"637c99bbfe115289cfedfb44","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/637c99bbfe115289cfedfb44/p4uSY0TKufJfcHpvEb_ZQ.jpeg","isPro":false,"fullname":"ssz","user":"ssz1111","type":"user","name":"ssz1111"},"summary":"Scientific figures are among the most effective means of communicating complex research ideas, yet producing publication-quality illustrations remains one of the most labor-intensive parts of paper preparation. Existing automated systems each target a single figure type under text-only input, leaving the diversity of types and conditions researchers actually use unaddressed; their raster outputs further cannot be locally revised. Because scientific figures are structured compositions of discrete semantic components, the localized errors generators produce on such layouts demand not a stronger backbone but a harness. We instantiate this harness in two complementary systems: Crafter, a multi-agent harness for figure generation that generalizes across figure types and input conditions without architectural changes, and CraftEditor, which applies the same pattern to convert raster outputs into editable SVGs. Moreover, we introduce CraftBench, a benchmark spanning three figure types and four input conditions with human quality annotation. Experiments show that Crafter substantially outperforms both standalone generators and the agentic baseline on PaperBanana-Bench and CraftBench, with ablations confirming each component's independent contribution; CraftEditor faithfully converts outputs into editable SVGs that surpass all baselines. Our code and benchmark are available at https://github.com/HaozheZhao/Crafter.","upvotes":54,"discussionId":"6a1e3aa9808ddbc3c7d43bca","githubRepo":"https://github.com/HaozheZhao/Crafter","githubRepoAddedBy":"user","ai_summary":"Automated systems for generating scientific figures face limitations in handling diverse figure types and conditions, prompting the development of multi-agent frameworks that generalize across different input scenarios and produce editable output formats.","ai_keywords":["multi-agent harness","figure generation","raster outputs","editable SVGs","CraftBench","PaperBanana-Bench"],"githubStars":1},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"637c99bbfe115289cfedfb44","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/637c99bbfe115289cfedfb44/p4uSY0TKufJfcHpvEb_ZQ.jpeg","isPro":false,"fullname":"ssz","user":"ssz1111","type":"user"},{"_id":"69bcecda3118c13015831d90","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/rRbbaM0n6SAvXxG5VQxJ_.jpeg","isPro":false,"fullname":"雨田梁","user":"victoriawhite7","type":"user"},{"_id":"69bcfc31157bf58b13af4b86","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/FYNpXgxZI4tkRj2HcG2Gw.jpeg","isPro":false,"fullname":"Siyu Luo","user":"aidentorres","type":"user"},{"_id":"650bba4e11752eec3074408f","avatarUrl":"/avatars/840c0638e655726af86019abde2d09b4.svg","isPro":false,"fullname":"Zhixiang Liang","user":"ZhixiangLiang","type":"user"},{"_id":"69bd40d275554b9ace435005","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/0qzeldPuZee5LrDF4UcHE.png","isPro":false,"fullname":"山崎凛","user":"ellacaif","type":"user"},{"_id":"698f8f12e7a7f2b4ed5a1774","avatarUrl":"/avatars/4542477cfa0c794be555831ba20e84f0.svg","isPro":false,"fullname":"Cydkauktam4ys","user":"cydkauktam4ys","type":"user"},{"_id":"69af7d90164b3dcc95c96cdf","avatarUrl":"/avatars/7fed3d8a2124910bef30fb7df9140422.svg","isPro":false,"fullname":"kak","user":"Kaowai","type":"user"},{"_id":"6984da8e9cda60784cd3ae53","avatarUrl":"/avatars/fe742e4f65f0724161f054bd95534c2f.svg","isPro":false,"fullname":"Emily Davis","user":"seckinkmchiroen5","type":"user"},{"_id":"69a3a3ea54551aa754f075fe","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/DyWOdnfe3WQJF_HxWySYY.jpeg","isPro":false,"fullname":"Попов Тимофей","user":"benjaminmitchel","type":"user"},{"_id":"699edef2ae7a66592b997d12","avatarUrl":"/avatars/51b0005e7b306f125d41a0e915e5c60c.svg","isPro":false,"fullname":"高晨曦","user":"hanhaoy","type":"user"},{"_id":"687363d49a81c7dcbcfa2d84","avatarUrl":"/avatars/5d943a5c811ed931c3fdcfee19253049.svg","isPro":false,"fullname":"jj","user":"realman123","type":"user"},{"_id":"6a159cbf6b16aa4c3df855ee","avatarUrl":"/avatars/31fc3437b009c37695be5c1ab532d29a.svg","isPro":false,"fullname":"Yang Zixuan","user":"yang-zixuan356","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":1,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.30611.md"}">

Papers

arxiv:2605.30611

Crafter: A Multi-Agent Harness for Editable Scientific Figure Generation from Diverse Inputs

Published on May 28

· Submitted by

ssz on Jun 2

#1 Paper of the day

Upvote

Authors:

Abstract

Automated systems for generating scientific figures face limitations in handling diverse figure types and conditions, prompting the development of multi-agent frameworks that generalize across different input scenarios and produce editable output formats.

AI-generated summary

Scientific figures are among the most effective means of communicating complex research ideas, yet producing publication-quality illustrations remains one of the most labor-intensive parts of paper preparation. Existing automated systems each target a single figure type under text-only input, leaving the diversity of types and conditions researchers actually use unaddressed; their raster outputs further cannot be locally revised. Because scientific figures are structured compositions of discrete semantic components, the localized errors generators produce on such layouts demand not a stronger backbone but a harness. We instantiate this harness in two complementary systems: Crafter, a multi-agent harness for figure generation that generalizes across figure types and input conditions without architectural changes, and CraftEditor, which applies the same pattern to convert raster outputs into editable SVGs. Moreover, we introduce CraftBench, a benchmark spanning three figure types and four input conditions with human quality annotation. Experiments show that Crafter substantially outperforms both standalone generators and the agentic baseline on PaperBanana-Bench and CraftBench, with ablations confirming each component's independent contribution; CraftEditor faithfully converts outputs into editable SVGs that surpass all baselines. Our code and benchmark are available at https://github.com/HaozheZhao/Crafter.