Advanced JSON Prompting: 12 Cases for Infographics, Creative Generation & Multi‑Round Iteration

Hello, I’m James. This follow‑up article dives deeper into using JSON‑structured prompts to solve three advanced use‑cases that plain natural‑language prompts struggle with: precise infographics, creative style‑fusion generation, and stable multi‑round iteration.

01 Infographics: Structure Determines Success

Pain point: When trying to create an AI‑generated "AI tool comparison" infographic with a 300‑word prompt, the model often drops columns, misplaces data, or produces a chaotic layout.

Root cause: Infographics require exact "data + structure + relationship". Natural language leaves the relationship to guesswork, while JSON fixes the relationships explicitly.

Case 1 – Horizontal Comparison Chart

Scenario: A product manager needs a competitor‑comparison infographic for a presentation.

{ "type": "横向对比信息图", "title": "2026 年 AI 助手三强对比", "subtitle": "速度 / 价格 / 能力 三维度拆解", "layout": "三列卡片式对比，顶部标题区，底部结论区", "columns": [ { "brand": "ChatGPT", "logo_color": "#10A37F", "speed": "⚡⚡⚡⚡", "price": "$$", "strength": "多模态 / 插件生态" }, { "brand": "Claude", "logo_color": "#D4622A", "speed": "⚡⚡⚡", "price": "$$$", "strength": "长文本 / 代码生成" }, { "brand": "Gemini", "logo_color": "#4285F4", "speed": "⚡⚡⚡⚡⚡", "price": "$$", "strength": "搜索整合 / 实时信息" } ], "conclusion": "综合评分：ChatGPT 全能 / Claude 深度 / Gemini 速度", "style": { "background": "浅灰白，卡片有轻微阴影", "typography": "中文标题 + 英文品牌名", "color_scheme": "各品牌色为主，灰色为辅" }, "constraints": "文字必须清晰可读，数据位置不得错乱，严格三列对齐" }

Result: Three neatly aligned brand cards with distinct colors, ready for direct PPT insertion.

Case 2 – Vertical Flowchart

Scenario: A technical team wants a "large‑model inference process" educational infographic.

{ "type": "竖向流程信息图", "title": "大模型是怎么「思考」的？", "layout": "从上到下，6 个步骤，步骤间用箭头连接，奇数步骤左对齐，偶数步骤右对齐，形成 Z 字型视觉流", "steps": [ {"no": 1, "label": "输入 Token", "icon": "键盘图标", "desc": "你打的每个字被切成 token"}, {"no": 2, "label": "Embedding 编码", "icon": "向量箭头", "desc": "token 变成高维向量"}, {"no": 3, "label": "注意力计算", "icon": "聚焦放大镜", "desc": "模型「看」每个词的关联"}, {"no": 4, "label": "多层变换", "icon": "堆叠层", "desc": "96 层 Transformer 逐层处理"}, {"no": 5, "label": "概率采样", "icon": "骰子", "desc": "从候选词中按概率选下一个词"}, {"no": 6, "label": "输出文本", "icon": "气泡框", "desc": "拼接所有词，形成你看到的回复"} ], "style": { "background": "深蓝渐变", "accent_color": "电光蓝 + 青绿", "font": "科技感无衬线字体" }, "constraints": "Z 字流程必须清晰，箭头方向正确，中文文案精准显示" }

Result: A Z‑shaped visual flow with deep‑blue tech colors, suitable for technical social media sharing.

Case 3 – Horizontal Timeline

Scenario: Creating an "AI history" timeline for a public‑account cover.

{ "type": "横向时间轴信息图", "title": "AI 40 年：从感知机到 GPT-4", "layout": "横向时间轴，主轴居中，关键事件上下交错标注", "timeline": [ {"year": "1986", "event": "反向传播算法", "position": "上"}, {"year": "1997", "event": "Deep Blue 击败卡斯帕罗夫", "position": "下"}, {"year": "2012", "event": "AlexNet 开启深度学习时代", "position": "上"}, {"year": "2017", "event": "Transformer 架构发布", "position": "下"}, {"year": "2022", "event": "ChatGPT 发布，全球爆红", "position": "上"}, {"year": "2026", "event": "GPT Image 2 发布", "position": "下"} ], "style": { "background": "米白色，轻微纸张纹理", "timeline_color": "深橙", "font": "中英混排，清晰可读" }, "constraints": "时间轴必须水平，年份标注准确，上下交错排列不重叠" }

Result: A clear horizontal timeline with alternating labels, gentle off‑white background for easy reading.

02 Creative Generation: JSON as a Style Palette

Pain point: When prompting for a "cyberpunk + ink‑wash" portrait, the model tends to pick only one style.

JSON solution: Split styles into primary_style and secondary_style and control their proportion with blend_ratio.

Case 4 – Style‑Fusion Portrait

Scenario: A designer needs a unique personal‑brand avatar that mixes cyberpunk and Chinese ink‑wash.

{ "type": "艺术融合人像", "subject": "年轻女性，短发，面朝正前方，表情淡然", "style": { "primary_style": "赛博朋克", "secondary_style": "中国水墨画", "blend_ratio": "6:4（赛博主导，水墨为底）", "primary_elements": "霓虹灯反光、电路线条、机械感细节", "secondary_elements": "墨晕笔触、留白空间、山水意境背景" }, "color": { "palette": "霓虹粉 + 电光蓝 + 水墨黑灰", "background": "深黑渐变，局部有墨迹晕染" }, "composition": "竖版人像，人物占画面 70%，背景虚化", "constraints": "不要出现文字，风格融合要自然，不要生硬拼接" }

Result: A neon‑bright figure against a subtle ink‑wash backdrop, creating a high‑impact East‑West aesthetic.

Case 5 – Concept Poster Variants

Scenario: A brand wants three different concepts for a "Human‑Technology Symbiosis" poster.

Technique: Use the variant field to define three separate creative directions.

{ "type": "概念主题海报", "concept": "人与科技共生", "variant": "A", "visual_metaphor": "人类的手与机械手交握，手心处生长出发光的植物", "mood": "温暖、希望、未来感", "color": { "background": "深夜蓝", "accent": "暖金色 + 生命绿", "texture": "轻微颗粒感胶片质感" }, "layout": "竖版 2:3，视觉中心在画面中央偏上", "typography": { "main_text": "共生", "sub_text": "与AI一起成长", "font_style": "中文衬线体，金色，右下角" }, "constraints": "两只手必须清晰，植物生长方向向上，文字可读" }

Result: A golden‑glowing plant emerging from clasped hands, conveying hope and future‑oriented technology.

Case 6 – Data‑Driven Creative Graphic

Scenario: A data team wants to visualise user‑retention metrics (D1 60 %, D7 30 %, D30 10 %) with a striking visual metaphor.

{ "type": "数据艺术化信息图", "title": "用户留存：消失的水滴", "concept": "用水滴蒸发的隐喻呈现用户留存衰减", "data": [ {"label": "D1", "value": "60%", "visual": "完整水滴，饱满"}, {"label": "D7", "value": "30%", "visual": "半蒸发水滴，有飘散效果"}, {"label": "D30", "value": "10%", "visual": "几乎蒸发，只剩底部水痕"} ], "layout": "横向排列三个水滴，从左到右逐渐消失，下方显示数据标签", "style": { "background": "深海蓝 #0A1628", "water_color": "冰蓝色，有光泽感", "evaporation": "白色烟雾状粒子效果" }, "constraints": "水滴大小比例必须符合数据比例，文字标签清晰，隐喻直观易懂" }

Result: The evaporating‑droplet metaphor makes the retention curve ten times more memorable than a plain line chart.

03 Multi‑Round Iteration: One JSON Skeleton, Stable Variants

The most powerful aspect of JSON prompting is the ability to keep a base template unchanged while only editing the fields that need variation.

Case 7 – Brand Festival Posters

Scenario: A designer must produce a series of 12 seasonal brand posters with a unified style.

{ "type": "节气品牌海报", "base_layout": "竖版 2:3，顶部 logo，中部主视觉，底部文案", "brand": { "name": "茶颜悦色", "primary_color": "#8B6914", "logo_position": "顶部居中" }, "style": { "overall": "新中式插画风", "background": "米白宣纸质感", "illustration_style": "工笔线描，淡彩填色" }, "current_festival": "谷雨", "main_visual": "细雨中的茶园，茶叶新芽，燕子飞过", "copy": { "main": "谷雨时节，新茶上市", "sub": "一杯春意，与君共饮" }, "constraints": "保持品牌色不变，中文文案精准显示，插画风格统一" }

Iterating only the current_festival and related visual fields yields a new poster while preserving layout, colors, and illustration style.

Case 8 – E‑Commerce Product Image A/B Test

Scenario: An e‑commerce operator wants to test which background color drives higher conversion.

{ "type": "电商产品主图", "product": "北欧风陶瓷咖啡杯，磨砂质感，哑光白色", "variant": "A", "background": { "color": "纯白 #FFFFFF", "texture": "无纹理", "props": "无" }, "lighting": "均匀柔光，无阴影", "composition": "杯子居中，俯视 45°，留白 30%", "constraints": "产品必须清晰，无多余元素，适合白底主图" }

{ "type": "电商产品主图", "product": "北欧风陶瓷咖啡杯，磨砂质感，哑光白色", "variant": "B", "background": { "color": "暖木色桌面", "texture": "原木纹理，轻微磨损感", "props": "旁边放一本翻开的书，一片叶子" }, "lighting": "暖黄自然光，从左侧射入，产品有柔和阴影", "composition": "杯子偏右，三分法构图，背景虚化", "constraints": "产品必须清晰，道具不抢镜，暖色系统一" }

Only the background and lighting fields change, keeping the product identical, which makes the A/B test variables pure and the results reliable.

Case 9 – Seasonal Character Design Series

Scenario: A game studio needs four seasonal outfits for the same character without altering the character’s base appearance.

{ "type": "角色设计图", "character": { "base": "20 岁东方女性，丸子头，杏眼，高颧骨，纤细身材", "expression": "淡然微笑", "pose": "站立，侧身 3/4 朝向右方" }, "season_variant": "春", "outfit": { "style": "改良汉服，轻薄飘逸", "color": "嫩绿 + 粉白", "details": "袖口绣桃花纹，腰间系浅绿绸带" }, "background": "模糊樱花树，浅粉色", "style": "游戏立绘风格，高饱和，清晰线条", "constraints": "角色面貌特征必须与基础描述一致，不得更改" }

Changing only season_variant and outfit produces four consistent seasonal renders ready for UI implementation.

04 Advanced Tips: Four Fields That Make JSON More "Obedient"

Constraint field with prohibitions : e.g., "constraints": "禁止出现文字以外的随机元素，禁止改变品牌色，禁止添加水印". The model respects these hard rules over natural‑language instructions.

Priority field to rank importance : e.g., "priority": "文字可读性 > 视觉美感 > 细节完整度". When elements conflict, the model follows the declared order.

Reference_style field for style cues : e.g., "reference_style": "类似《WIRED》杂志封面的技术感配色，不要复制，只要同类气质". Provides precise stylistic direction.

Iteration_note field to record the previous version's issue : e.g., "iteration_note": "上一版背景太暗，主体不突出，这次请加强主体光源". Embedding the feedback directly in JSON improves consistency across rounds.

05 Comprehensive Templates: Full‑Featured JSON Advanced Blueprint

Case 10 – Data Report Cover

{ "type": "报告封面", "report": { "title": "2026 中国 AI 产业年度白皮书", "issuer": "XX 咨询研究院", "date": "2026 年 Q2" }, "main_visual": "俯视中国地图轮廓，各省市有光点连线，形成神经网络感", "color": { "background": "深海蓝 #0A1628", "accent": "电光蓝 + 白色", "map_glow": "蓝白渐变发光" }, "layout": { "format": "竖版 A4 比例", "title_position": "左下角，白色大字", "visual_coverage": "全幅背景，地图占 60%" }, "priority": "标题文字可读性 > 地图视觉效果", "constraints": "报告名称必须完整清晰，地图轮廓必须准确，不要出现无关图标" }

Case 11 – Brand Annual‑Report Data‑Art Fusion

{ "type": "品牌数据艺术图", "headline_data": "用户增长 300%", "visual_concept": "三棵树，从左到右越来越高大，树干是品牌主色，树叶是用户头像拼贴", "layout": "横版 16:9，三棵树从左到右排列，下方标注增长数据", "brand": { "name": "DEMO 品牌", "primary_color": "#FF6B35", "logo": "右上角小 logo 区域留白" }, "data_labels": [ {"year": "2024", "tree": "第1棵", "users": "100 万"}, {"year": "2025", "tree": "第2棵", "users": "200 万"}, {"year": "2026", "tree": "第3棵", "users": "400 万"} ], "style": "品牌插画风，扁平化，温暖橙调", "constraints": "三棵树高度比例必须与数据比例对应，文字标注清晰" }

Case 12 – Multi‑Round Iteration Ultimate: Character × Scene × Emotion Matrix

{ "type": "影视预视化图", "character_fixed": { "desc": "40 岁男性，银发，深邃眼神，长风衣", "note": "此字段全部 9 张图保持完全一致" }, "scene": "[填入：荒野 / 都市 / 密室]", "emotion": "[填入：愤怒 / 绝望 / 平静]", "style": "电影级写实摄影，柯达 Portra 400 胶片感", "constraints": "角色特征不变，scene 和 emotion 控制画面氛围和环境" }

By looping over the scene and emotion fields, nine distinct visuals are generated from a single skeleton, demonstrating the efficiency of structured iteration.

Conclusion

Infographic scenarios benefit from separating data, structure, and relationships into dedicated JSON fields.

Creative generation uses primary_style, secondary_style, and blend_ratio to precisely control style fusion.

Multi‑round iteration locks the stable skeleton and edits only variable fields, guaranteeing series consistency.

The trio of constraints, priority, and iteration_note are key to making models follow instructions reliably.

A/B testing and series production are the real productivity use‑cases where JSON outperforms free‑form prompts.