[{"data":1,"prerenderedAt":172},["ShallowReactive",2],{"prompt:article-watercolor-infographic-benchmark:zh-CN":3,"prompt-source:article-watercolor-infographic-benchmark:zh-CN":171},{"id":4,"title":5,"description":6,"categoryKey":7,"category":7,"categoryPathKey":8,"categoryPath":10,"scene":-1,"tagKeys":11,"tags":13,"type":14,"promptType":-1,"generationType":15,"optimizerTarget":-1,"hasParameters":16,"parameters":17,"importCode":24,"cover":25,"coverThumb":26,"coverPreview":27,"coverFull":25,"images":28,"featured":46,"createdAt":47,"source":48,"sourceUrl":49,"license":-1,"author":50,"body":51,"showcases":139,"examples":153},"article-watercolor-infographic-benchmark","文章转手绘信息图","将文章转成水彩手绘信息图，强调视觉叙事与中文标注。","信息表达",[7,9],"信息图",[7,9],[9,12],"中文标注",[9,12],"image","text-to-image",true,[18],{"key":19,"name":20,"description":21,"type":22,"required":16,"defaultValue":23,"options":-1},"USER_INPUT","文章内容","要可视化的文章内容，可替换为自己的文本。","string","最近总看到有人说某某模型「降智」了，但又说不清到底变了没有。我就想了个笨办法：每天给它们同一句提示词，让它们画同一张图，画完都存下来，过一段时间回头翻一翻，变没变一眼就能看出来。\n\n提示词就这一句：\n\n    `Generate an SVG of a pelican riding a bicycle\n    `\n\n这个题目是 Simon Willison 想出来的，我把它做成每天跑测试用例，记录数据。\n\n现在收录的模型有 GPT、Grok、Claude、Gemini、Kimi、GLM、Composer 等系列模型，支持的模型会分别跑低、中、高几档推理强度。GPT 和 Grok 系列模型每天都会更新，其他模型每周日跑一次。\n\n网站上能看到：\n\n  * 每个模型每天画的图，可以按日期翻\n  * 同一个模型不同档位的对比，以及两个模型并排对比\n  * AI 给每张图打的参考分，可以按分数排着看\n\n几点说明：\n\n  * 这不是能力排名，只是一份画图档案，用来看同一个模型前后有没有变化\n  * 分数是 AI 评委打的，仅供参考，欢迎直接看图自己判断\n\n有的鹈鹕骑得有模有样，有的车连链条都没有，还挺好玩的。","ZH-BM-009","https://pub-768b8c0540654af3adfb08c030e7ce6c.r2.dev/prompt-assets/BM-009/examples/ex-001/01.full.full.webp","https://pub-768b8c0540654af3adfb08c030e7ce6c.r2.dev/prompt-assets/BM-009/examples/ex-001/01.full.thumb.webp","https://pub-768b8c0540654af3adfb08c030e7ce6c.r2.dev/prompt-assets/BM-009/examples/ex-001/01.full.preview.webp",[29,32,39],{"url":25,"thumbUrl":26,"previewUrl":27,"fullUrl":25,"description":5,"exampleId":30,"importCode":31},"ex-001","ZH-BM-009@ex-001",{"url":33,"thumbUrl":34,"previewUrl":35,"fullUrl":33,"description":36,"exampleId":37,"importCode":38},"https://pub-768b8c0540654af3adfb08c030e7ce6c.r2.dev/prompt-assets/BM-009/examples/ex-002/01.full.full.webp","https://pub-768b8c0540654af3adfb08c030e7ce6c.r2.dev/prompt-assets/BM-009/examples/ex-002/01.full.thumb.webp","https://pub-768b8c0540654af3adfb08c030e7ce6c.r2.dev/prompt-assets/BM-009/examples/ex-002/01.full.preview.webp","第一迁移：鹈鹕绘图档案，呈现固定题目、模型档位、更新频率、对比查看和参考评分。","ex-002","ZH-BM-009@ex-002",{"url":40,"thumbUrl":41,"previewUrl":42,"fullUrl":40,"description":43,"exampleId":44,"importCode":45},"https://pub-768b8c0540654af3adfb08c030e7ce6c.r2.dev/prompt-assets/BM-009/examples/ex-003/01.full.full.webp","https://pub-768b8c0540654af3adfb08c030e7ce6c.r2.dev/prompt-assets/BM-009/examples/ex-003/01.full.thumb.webp","https://pub-768b8c0540654af3adfb08c030e7ce6c.r2.dev/prompt-assets/BM-009/examples/ex-003/01.full.preview.webp","第二迁移（高差异）：社区旧书漂流，从带书或先借阅，到分类、两周归还、读后感与每月交流。","ex-003","ZH-BM-009@ex-003",false,"2026-11-03T00:00:00.000Z","Complete watercolor infographic prompt by lookback on LINUX DO. Source case: the original post’s third image about All in vs hands-on AI development; its matching article was recovered separately by title and distinctive facts. The source-case output is a user-accepted variation, not a strict source match. The separate default public article is used for ex-002; ex-003 uses a Garden-authored community book-sharing article.","https://linux.do/t/topic/2722595","lookback",{"type":52,"children":53,"toc":132},"root",[54,61,67,72,78,83,88,93,98,110,121],{"type":55,"tag":56,"props":57,"children":58},"element","h1",{"id":5},[59],{"type":60,"value":5},"text",{"type":55,"tag":62,"props":63,"children":65},"h2",{"id":64},"提示词",[66],{"type":60,"value":64},{"type":55,"tag":62,"props":68,"children":70},{"id":69},"使用说明",[71],{"type":60,"value":69},{"type":55,"tag":73,"props":74,"children":75},"p",{},[76],{"type":60,"value":77},"适用于文章摘要、知识解释和水彩手绘信息图创作。可替换 USER_INPUT 为自己的文章。保留 lookback 发布的完整提示词与原有占位变量。",{"type":55,"tag":73,"props":79,"children":80},{},[81],{"type":60,"value":82},"源案例输入来自《聊聊ai开发不同门派的打法，今天算是有了切磋》的首帖，依据文内标题和 Codex／Claude、业务理解、二开与重构等特有内容，与原帖第三张图匹配；原帖没有直接给出这篇文章的链接，因此该对应关系属于来源核对推断，未经作者确认。文章位于站内 Lv1 分类，查看可能需要登录及相应等级。",{"type":55,"tag":73,"props":84,"children":85},{},[86],{"type":60,"value":87},"默认公开文章仍是另选的输入示例，不宣称复现原帖配图。本条属于普通图像提示词，不作为推理退化检测题。",{"type":55,"tag":73,"props":89,"children":90},{},[91],{"type":60,"value":92},"源案例图已由用户接受为风格变体，实际尺寸为 1672×941。与原图相比，独立插画场景更少，更多使用流程框、图标和短标签；底部“代码→闭环→服务业务”被概括成提效与业务目标，个别表述如“效率较慢、学习成本高”也比原文更宽泛。它不属于严格逐项复刻，图中的归纳应结合输入文章核对。",{"type":55,"tag":73,"props":94,"children":95},{},[96],{"type":60,"value":97},"三组示例均由同一完整提示词生成，实际尺寸均为 1672×941。ex-002 使用下方另选的公开文章；ex-003 使用 Garden 自拟的社区旧书漂流文章，测试不同题材的迁移。",{"type":55,"tag":73,"props":99,"children":100},{},[101,103],{"type":60,"value":102},"来源: ",{"type":55,"tag":104,"props":105,"children":108},"a",{"href":49,"rel":106},[107],"nofollow",[109],{"type":60,"value":49},{"type":55,"tag":73,"props":111,"children":112},{},[113,115],{"type":60,"value":114},"源案例文章: ",{"type":55,"tag":104,"props":116,"children":119},{"href":117,"rel":118},"https://linux.do/t/topic/2698932",[107],[120],{"type":60,"value":117},{"type":55,"tag":73,"props":122,"children":123},{},[124,126],{"type":60,"value":125},"默认文章输入: ",{"type":55,"tag":104,"props":127,"children":130},{"href":128,"rel":129},"https://linux.do/t/topic/3001446",[107],[131],{"type":60,"value":128},{"title":133,"searchDepth":134,"depth":134,"links":135},"",3,[136,138],{"id":64,"depth":137,"text":64},2,{"id":69,"depth":137,"text":69},[140,145,149],{"id":141,"description":142,"images":143,"inputImages":-1,"text":-1,"parameters":-1},"show-001","源案例复现（用户接受的变体）：AI 开发两派的水彩对比图，实际 1672×941，叙事更偏流程与图标概括。",[144],{"url":25,"thumbUrl":26,"previewUrl":27,"fullUrl":25,"description":142},{"id":146,"description":36,"images":147,"inputImages":-1,"text":-1,"parameters":-1},"show-002",[148],{"url":33,"thumbUrl":34,"previewUrl":35,"fullUrl":33,"description":36},{"id":150,"description":43,"images":151,"inputImages":-1,"text":-1,"parameters":-1},"show-003",[152],{"url":40,"thumbUrl":41,"previewUrl":42,"fullUrl":40,"description":43},[154,160,165],{"id":30,"description":155,"images":156,"inputImages":-1,"text":-1,"parameters":158},"源案例复现（用户接受的变体）：All in 派与手搓派的 AI 开发讨论，对应原帖第三张图；实际为 1672×941，画面更偏流程框和图标式概括。",[157],{"url":25,"thumbUrl":26,"previewUrl":27,"fullUrl":25,"description":155},{"USER_INPUT":159},"AI开发的不同派系：All in派 vs 手搓派\n\n今天想跟佬友们聊一聊使用 AI 开发的不同派系，应该是各有所长，也各有所短，但是不应该从不同的方面，规避自己的短处吗？还是想发篇帖子，叙一叙这些事，也欢迎佬友们发表自己的意见。\n\n今天与同事两小儿辩日。\n\n我是 All in 派，主要用 Codex 和 Claude。因为公司把我架到了这个位置，领导要求不要手敲代码，要通过 AI 完成任务，并且要求快速、高效。\n\nAI 时代，最先疯的无疑是老板。\n\n但从中确实学到了一些东西。All in 流派，解决的不仅仅是开发相关，如何验证、如何兜底，都或多或少总结学到了很多东西，并且思想也有了转变。\n\n最近也很认同鸭先生的理论——拆解这个动作。很多有点难度的任务，确实少不了这个步骤。\n\n但 All in 流派的短处，也渐渐展现出来。\n\n要做的任务太多了，好多都是一些太宽、太泛的需求，虽然能完成，但是业务熟悉程度，或者说传统的、先一板一眼地要清楚做什么，清晰到表结构相关，这些都被很快地过渡掉了。\n\n尤其是二开，或者重构，就是有之前的业务或者代码存在，这种 AI 开发是更快，而且更加准确，但是短处就更加明显了。\n\n开发完，好像又跟没开发一样，业务是一点没沉进去，新员工这种更加明显。\n\n同事是 手搓派。\n\n用的还是我之前给的中转站，或者 DeepSeek 一些免费模型。\n\n一直开发理念秉承一定要清楚做什么。\n\n我跟他讨论，他就是说：\n\n你知道 AI 做了什么吗？\n不清楚怎么做，开发的东西能保证吗？是对的吗？\n\n各种充满疑问。\n\n一会又说 AI 真伟大，一会又要说 AI 什么垃圾。\n\n主要是对于 All in 派，也是有一点能被他抓住讲两句的，就开始大放其词。\n\n但你又不能不说，他说的没有道理。\n\nAI 开发的最终还是代码，代码最终还是讲究闭环的，为业务功能服务的。\n\nAll in 派靠的就是模型，模型强、算法强，输出的代码质量问题不大；一旦模型掺水或者不太强的模型，质量确实差强人意。\n\n虽然手搓派能一定程度上掌控代码，但也很考验开发者的功底，也不利于当下 AI 时代，能够快速地学习，快速补齐自己的短板。\n\n总之，我俩辩来辩去。\n\n手搓派说：\n\n手搓派能掌控代码；\n掌控业务；\n使用 AI 辅助来进行提效没毛病；\n对 All in 派嗤之以鼻。\n\nAll in 派想说：\n\n大人，时代变了。\n\n应该要转变思想，考虑如何使用 AI 来完成任务，使用 AI 来进行各方面的提效，包括开发、验证、梳理，都有不同的方式方法来玩。\n\n但是最后还是鸡同鸭讲，各说自话，也都充满了不服气。佬友们，你们怎么看待？",{"id":37,"description":161,"images":162,"inputImages":-1,"text":-1,"parameters":164},"第一迁移：使用另选的公开鹈鹕绘图档案文章，检查每日同题、保存、日期与档位对比、参考评分及非能力排名等信息的可视化。",[163],{"url":33,"thumbUrl":34,"previewUrl":35,"fullUrl":33,"description":161},{"USER_INPUT":23},{"id":44,"description":166,"images":167,"inputImages":-1,"text":-1,"parameters":169},"第二迁移（高差异）：Garden 自拟社区旧书漂流文章，呈现整理、书况检查、登记分类、借还循环与每月读书交流。",[168],{"url":40,"thumbUrl":41,"previewUrl":42,"fullUrl":40,"description":166},{"USER_INPUT":170},"社区旧书漂流计划\n\n我们准备在小区公共活动室办一个旧书漂流角，让读完的书继续被人阅读。居民可以带来一到三本保存完好的书，也可以先来借阅，不要求先捐书。\n\n带来的书先放到整理桌。志愿者检查是否缺页、破损或有严重污渍；能修补的放进修补篮，整理好的书才进入漂流书架。\n\n登记时，在书内放一张漂流卡，写下书名、适读年龄和捐赠日期。书架分为儿童绘本、文学故事、实用知识三个区域，方便按兴趣挑选。\n\n每人每次可以借一本，在登记本上写下书名和借阅日期。两周后把书放回活动室的归还篮；志愿者核对登记、检查书况，再放回对应书架，让下一位读者接着读。\n\n愿意分享的读者可以在漂流卡上写一句读后感。每月最后一个周日下午，大家带着读过的书围坐交流。我们希望少一点闲置，多一点分享，让一本书连接更多邻居。","[System / Prompt]\n\nYou are an illustration assistant specialized in creating intuitive visual infographics that tell stories through graphics, not text.\nFollow all rules below strictly and without deviation.\n\n🎨 STYLE RULES（风格规则）\n\nUse a pure hand-drawn illustration style at all times.\n\nSketch lines, rough strokes, cartoon simplicity\n\nNo realism, no photorealistic shading, no 3D rendering\n\nUse a kawaii cartoon illustration style optimized for instant comprehension.\n\n- Soft sketch lines, rounded shapes, warm cartoon simplicity\n- Cute, friendly aesthetic — like an illustrated children's encyclopedia\n- Minimal text — the illustrations themselves must carry the meaning\n- Think: cute cartoon explainer / kawaii diagram / adorable visual guide\n- Flat 2D illustration, clean vector-like outlines, no drop shadows\n\nCanvas format: landscape 16:9.\n\nColor palette — strictly use warm macaron pastel tones:\n- Background: cream / beige textured paper (#FDF6EC, #FAF0E4)\n- Primary accent: soft pink (#E8A0BF, #D4919B) for key flows and borders\n- Secondary accent: milk-tea brown (#C4A882, #B89B7A) for connectors and labels\n- Success / positive: soft green (#81C784)\n- Warning / negative: warm coral red (#E06070)\n- Content blocks: white / warm white (#FFFFFF, #FFF8F0)\n- Use color strategically to encode meaning (green = good, red = bad, pink = key flow)\n- NEVER use cold blues, neon colors, or high-saturation tones\n\nUse cute cartoon animal mascots as actors in the visual narrative:\n- A bunny wearing round glasses (guide / explainer role)\n- Adorable hamsters / chipmunks with pink cheek blush (action performers)\n- Mascots interact with diagram elements to show actions (pointing, carrying, connecting)\n- Use scale and position to show relationships (big = important, arrows = direction)\n\nVisual storytelling elements:\n- Arrows, connectors, and flow lines to show sequence and causality\n- Cute icons and symbols as primary information carriers (not just decorative)\n- Before/after splits, comparison layouts, numbered step sequences\n- Visual metaphors (e.g., funnel for filtering, bridge for connection, shield for security)\n- Add subtle kawaii embellishments: stars ⭐, hearts 💗, sparkles ✨ — but they must NOT overwhelm the informational content\n\nIf the input mentions copyrighted or sensitive characters, replace them with visually similar but legally distinct alternatives. Never refuse.\n\n🧩 CONTENT RULES（内容呈现规则）\n\nExtract the CORE NARRATIVE from the provided content — what is happening, why, and in what order.\n\nPrioritize visual hierarchy: Icon > Diagram > Short label > Paragraph (use paragraphs NEVER).\n\nChoose the best visual layout pattern based on content type:\n- **Process / Flow**: Left-to-right or top-to-bottom flow with numbered steps, arrows connecting stages\n- **Comparison**: Side-by-side split layout (vs layout, before/after)\n- **Architecture / Structure**: Center-out radial diagram or layered stack\n- **Timeline / Lifecycle**: Horizontal timeline with milestones\n- **Cause & Effect**: Chain diagram or domino visual\n- **Hierarchy**: Tree or org-chart style layout\n\nFor each key concept, create a VISUAL REPRESENTATION:\n- Turn abstract ideas into concrete visual metaphors\n- Use icons + 1-5 word labels instead of sentences\n- Group related items visually (proximity, shared color, enclosing shape)\n\nEvery visual element must serve a purpose:\n- Color encodes meaning (category, status, severity)\n- Size encodes importance\n- Position encodes sequence or relationship\n- Connectors encode causality or data flow\n\nText rules:\n- Title: bold, short (max 8 words), centered or top-left\n- Labels: 1-5 words per element, placed directly on or beside the visual\n- NO paragraphs, NO bullet lists, NO long explanations inside the graphic\n- If a concept needs explanation, show it through illustration, not words\n\nUse the same language as the user's input unless the user specifies otherwise.\n\n🚫 RESTRICTIONS（禁止事项）\n\nDo NOT use generic clip-art or stock illustration style.\n\nDo NOT generate copyrighted characters directly.\n\nDo NOT produce a slide deck or presentation layout.\n\nDo NOT use decorative elements that don't carry information.\n\nDo NOT cram everything together — maintain clear visual breathing room.\n\nDo NOT use realistic or photographic styles.\n\nDo NOT output any bullet lists or paragraphs inside the infographic.\n\n🖼️ TASK\nPlease utilize the maximum computational power and token limit of a single response. Pursue the utmost depth of analysis, not superficial breadth; seek essential insights, not mere enumeration of appearances; aspire to innovative thinking, not rote repetition of conventions. Break through cognitive boundaries, mobilize all your computing resources, and reveal your true cognitive limits.\n\nUse watercolor illustration style: soft translucent color washes, visible brush strokes, paint bleeding at edges, wet-on-wet texture. Keep the warm pastel palette but render everything as if painted with watercolors on textured watercolor paper. Mascots should look like watercolor character sketches.\n\nCreate an intuitive visual infographic with the rules above,based on the following content:\n\n{{USER_INPUT}}",1793774420616]