SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsimage-service

image-service

Design & creation

View repository

Updated 10-16 01:48

About this skill

zlab image processing skill

Configuration

Configuration Item Value
API 配置文件中设置
Image generation model 配置文件中设置
Vision model 配置文件中设置
Configuration file config/settings.json

Execution specifications

  • Script usageAbsolute pathcall, save the image tocurrent working directory
  • PromptMust use Chinese, title labels in the imagemust be in Chinese
  • when containing Chinese text, append the text clarity suffix from guides/text-rendering.md text within is clear suffix
  • Default aspect ratio 16:9, default for vertical scenarios (Xiaohongshu/comics) 3:4

🔀 Scenario routing (iron rule: first step after receiving a request)

收到生图需求
  │
  ├─ 信息图/数据可视化/架构图/流程图/对比图/鱼骨图/思维导图/金字塔/漏斗/韦恩图/冰山图
  │  → 📊 信息图 → scenes/infographic/
  │
  ├─ 封面/头图/banner/公众号封面/文章封面
  │  → 🖼️ 封面图 → scenes/cover/
  │
  ├─ 长图/微信长图/多屏/从上到下/竖版长图/分段/多个段落
  │  → 📜 长图 → scenes/long-image/
  │
  ├─ 幻灯片/PPT图/slides/演示文稿
  │  → 📑 幻灯片 → scenes/slide-deck/
  │
  ├─ 漫画/连环画/分镜/绘本/知识漫画/故事图
  │  → 📖 漫画 → scenes/comic/
  │
  ├─ 文章插图/自动配图/智能插图/给文章配图
  │  → ✏️ 文章插图 → scenes/article-illust/
  │
  ├─ 小红书/笔记配图/种草图/卡片系列
  │  → 📱 小红书图 → scenes/xhs/
  │
  ├─ 海报/主图/电商/促销/九宫格/产品图/营销/物料
  │  → 🛍️ 营销物料 → scenes/marketing/
  │
  ├─ 改图/编辑图片/风格转换/基于这张图
  │  → 🔄 图生图 → core/image_to_image.py（直接执行）
  │
  ├─ 分析图片/OCR/识别文字/描述图片/图生文
  │  → 👁️ 图生文 → core/image_to_text.py（直接执行）
  │
  └─ 其他单图需求（一张图，无特殊场景）
     → 🎨 单图 → core/text_to_image.py（直接执行）

After routing to a scenario → load that scenario's README.md to get detailed parameters and usage → operate according to the execution flow below.

📋 General execution flow (shared by all scenarios)

Step 1: Analyze requirements

  • Clarify what to generate (theme, content points)
  • Determine the scenario type (already determined by routing)
  • Determine the quantity (how many images)
  • Determine the ratio (16:9 / 3:4 / 1:1, etc.)

Step 2: Plan the solution (required for multi-image scenarios)

Single image: can be skipped and generated directly. Multi-image scenarios (long image/slides/comics/Xiaohongshu/article illustrations): must first output a plan and execute after user confirmation.

Plan output format:

| 序号 | 内容概要 | 类型/布局 |
|-----|---------|----------|
| 1   | xxx     | xxx      |
| 2   | xxx     | xxx      |

**场景**：xx | **风格**：xx | **比例**：xx | **预计**：N张

Step 3: Execute generation

Scenario Execution method Rules
📊 Infographic Scenario script called directly Single image can be concurrent
🖼️ Cover image Scenario script called directly Single image
📜 Long image core engine serial Iron rule: must be serial! First screen text_to_image, subsequent image_to_image references the previous screen, finally merge and stitch
📑 Slides Scenario script page by page Generate page by page serially
📖 Comics Scenario script page by page Serial! The first page text_to_image sets the tone, subsequent image_to_image references the previous page
✏️ Article illustrations Scenario script batch Each chapter is independent and can be concurrent (≤8 images)
📱 Xiaohongshu Scenario script image by image Serial! The first image sets the tone, subsequent ones reference the previous image to keep the style consistent
🛍️ Marketing materials core engine + template Same style can be concurrent (≤8 images)
🔄 Image-to-image core called directly Single image
🎨 Single image core called directly Single image

Step 4: Quality Check

  1. After generation is complete,images containing Chinese textUsing core/image_to_text.py -m ocr verify whether the text is clear
  2. OCR result does not match expectations → use core/image_to_image.py iterative fixes (refer to guides/text-rendering.md）
  3. After the multi-image scenario is complete, show them to the user one by one for confirmation, and regenerate any individual image the user is not satisfied with

🛠️ Script Quick Reference

Scenario Scripts

# 📊 信息图 — 20布局×17风格=340种
python3 scenes/infographic/zlab_infographic.py -l {布局} -s {风格} -n "标题" -c "内容" -o out.png
python3 scenes/infographic/zlab_infographic.py --list  # 查看所有选项

# 🖼️ 封面图 — 五维定制=3888种
python3 scenes/cover/zlab_cover.py -n "标题" --type {类型} --palette {配色} --rendering {渲染} --text {文字} --mood {氛围} -o out.png

# 📑 幻灯片 — 16种预设
python3 scenes/slide-deck/zlab_slide_deck.py article.md --style {预设} -o slides/
python3 scenes/slide-deck/zlab_slide_deck.py article.md --outline-only  # 仅看大纲

# 📖 漫画 — 5画风×7基调×6布局=210种
python3 scenes/comic/zlab_comic.py source.md --art {画风} --tone {基调} --layout {布局} -o comic/

# ✏️ 文章插图 — 6类型×8风格=48种
python3 scenes/article-illust/zlab_article_illustrator.py article.md --style {风格} -o images/

# 📱 小红书 — 9风格×6布局=54种
python3 scenes/xhs/zlab_xhs_images.py content.md --style {风格} --layout {布局} -o xhs/

All scenario scripts support --list to view available options.

Core Engine

# 文生图
python3 core/text_to_image.py "中文描述" -r 16:9 -o output.png
python3 core/text_to_image.py "描述" -r 3:4 --ref ref.png -o output.png  # 参考图风格

# 图生图
python3 core/image_to_image.py input.png "编辑描述" -r 3:4 -o output.png

# 图生文
python3 core/image_to_text.py image.jpg -m describe|ocr|chart|fashion|product|scene

# 长图拼接
python3 core/merge_long_image.py img1.png img2.png -o output.png --blend 20
python3 core/merge_long_image.py -p "*.png" -o output.png --sort name

Supported ratios:1:1 2:3 3:2 3:4 4:3 4:5 5:4 9:16 16:9 21:9

📚 General Guide (load as needed)

Trigger Conditions File When to Load
Image contains Chinese text guides/text-rendering.md When the prompt requires Chinese titles/labelsMust load
Create images for PPT/documents guides/color-sync.md Load when color coordination with other media is needed
Need to optimize image generation results guides/prompt-guide.md Load when image generation results are unsatisfactory and need improvement
API interface issues guides/api-reference.md Load when debugging the underlying interface
Generate multi-screen long images scenes/long-image/README.md When routing to the long-image scenarioMust load
Marketing material templates scenes/marketing/templates.md When routing to the marketing scenarioMust load