According to AlphaMoat, as of 2026-09, Infographic Image has 1,146 downloads and 0 stars — ranked #4,965 of 5,078 AI skills overall, and #501 of 512 in Design & Media.
| Platform | Installs | Downloads | Stars | Comments |
|---|---|---|---|---|
| ClawHub ↗ | 30 | 1.1K | - | - |
| SkillHub ↗ | 137 | 1.1K | - | — |
Platforms count differently and update at different times.
The skill accepts a topic, text, or file, selects a built-in visual style, and produces a structured image-generation prompt. It can preview the prompt without rendering or call an image model and save the result locally. The default style is a white magnetic whiteboard; comic, editorial, and spatial-gallery styles are also available. Source text may be sent to DashScope for inference or image generation, so do not process confidential documents without approval. Prompt payloads and images are written to user-selected or default local paths; check their contents and locations before sharing or reusing them.
Profile published 2026-10-29
Generated from the skill’s documentation as examples; not results of actual runs.
The counting method changed in 2026-09: skills with the same name are now counted separately by author, and the full catalog is covered. Dashed lines show months before the change; they are not compared with later months, and no growth is computed across the change.
| Month | Downloads | MoM | Stars |
|---|---|---|---|
| 2026-09 | 1.1K | — | - |
| 2026-08 | 1.0K | +10.7% | - |
| 2026-07 | 931 | +25.0% | - |
| 2026-06 | 745 | +16.4% | - |
| 2026-05 | 640 | — | - |
| 2026-03 | 390 | — | - |
The author’s original file, shown as plain text without edits or review by AlphaMoat. Links are not clickable; check commands before running them. Fetched 2026-10-23.
--- name: infographic-image description: >- 将任意主题、长文、报告、纪要或说明文本稳定转换成中文视觉生图提示词, 然后调用 DashScope Qwen 图像模型直接出图。适用于“把这段内容做成信息图”、 “做成故事漫画长图”、“长文转图”、“生成中文生图 prompt”、“根据文档出图” 等场景;默认采用白色磁吸会议板手写风格,也支持切换到其他内置模板。 --- # Infographic Image
## Overview 使用 `scripts/generate_infographic.py` 将任意输入内容压缩为稳定的中文视觉生图提示词,再调用 DashScope 默认模型 `qwen-image-2.0-pro` 出图并下载到本地。 ## Quick Start 1. 配置 API Key: ```bash export DASHSCOPE_API_KEY="<your_api_key>" ``` 2. 从长文直接生成图像: ```bash python3 scripts/generate_infographic.py \ --text-file /absolute/path/article.md \ --size "1440*1800" ``` 3. 只生成稳定 prompt,不出图: ```bash python3 scripts/generate_infographic.py \ --text "这里放任意中文或英文内容" \ --dry-run ``` 4. 加载自定义风格补充: ```bash python3 scripts/generate_infographic.py \ --text-file /absolute/path/report.txt \ --style-preset comic-story \ --style-file /absolute/path/custom-style.md ``` ## Workflow 1. 从 `--topic`、`--text` 或 `--text-file` 收集原始素材。 2. 从内置模板中选择一套协议与风格;若未指定,则默认使用白色磁吸会议板手写风格。 3. 可选叠加 `references/styles/*.md` 或外部风格文件作为补充。 4. 调用 DashScope 文本模型生成严格 JSON,提取 `image_generation_prompt`。 5. 使用默认模型 `qwen-image-2.0-pro` 渲染图像并保存到本地。 6. 若推理失败,使用脚本内置兜底模板生成可用 prompt。 ## Key Files - `references/meta-prompt.md`:信息图模板的元提示词。 - `references/meta-prompts-comic-story.md`:新中式漫画故事模板的元提示词。 - `references/styles/default.md`:信息图手绘笔记风格。 - `references/styles/clean-editorial.md`:信息图编辑排版风格。 - `references/styles/comic-story.md`:新中式漫画故事风格。 - `references/meta-prompts-spatial-gallery.md`:空间化画廊信息图模板的元提示词。 - `references/styles/spatial-gallery.md`:洁净实验室 / 画廊装置信息图风格。 - `references/meta-prompts-frosted-whiteboard.md`:白色磁吸会议板手写图模板的元提示词。 - `references/styles/frosted-whiteboard.md`:白色磁吸会议板 + 手写板书风格。 - `scripts/generate_infographic.py`:统一入口脚本。 ## Extend Styles 内置模板当前包含 `default`、`clean-editorial`、`comic-story`、`spatial-gallery`、`frosted-whiteboard`。其中 `default` 默认就是白色磁吸会议板手写风格;如需随机,可显式传 `--style-preset random`。新增模板时,优先补齐对应协议与风格文件;如果只是临时试验,也可以直接传 `--style-file` 或 `--style-hint`。 ## High-Value Options - `--topic`:简短主题。 - `--text`:直接传长文本。 - `--text-file`:从本地文件读取正文。 - `--style-preset`:选择内置模板,如 `default`、`clean-editorial`、`comic-story`、`spatial-gallery`、`frosted-whiteboard`;留空走默认,传 `random` 才随机。 - `--style-file`:加载外部风格文件。 - `--style-hint`:临时补充风格偏好。 - `--dry-run`:只输出推理出的 prompt 和负面提示词,不调用生图。 - `--prompt-output`:将推理结果 JSON 保存到本地。 - `--size`:输出尺寸,如 `1440*1800`、`1664*928`、`1328*1328`。 - `--prompt-model`:提示词推理模型,默认 `qwen-plus`。 - `--image-model`:可选覆盖生图模型;默认 `qwen-image-2.0-pro`,通常无需修改。 ## Reliability Rules 1. 长文本优先通过 `--text-file` 输入,避免 shell 转义污染。 2. 模板级规则放在 `references/meta-prompt*.md`,不要把长规则堆回 `SKILL.md`。 3. 风格扩展放进 `references/styles/`,保持工作流稳定、风格可替换。 4. 生成失败时先用 `--dry-run` 检查推理出的 prompt,再决定是否改风格补充。 5. 默认按 `qwen-image-2.0-pro` 使用自由宽高尺寸,建议直接传目标成图尺寸,如 `1440*1800` 或 `1600*1200`。
ContextWeave one-shot architecture diagram generation
A script-based skill that generates or edits images with Nano Banana Pro.
Runs the whisper command line to transcribe audio locally.
A skill that generates and edits diagrams through an MCP server.
Design intelligence and implementation guidance for polished interfaces, from layout and flows to tokens and frontend changes.
An AI video generation and orchestration skill across several model APIs.
Pull a frame out of a video with ffmpeg, or produce a thumbnail for a quick look.
Turn audio files into text with local Whisper, running fully offline once the model is downloaded.
Data month: 2026-09 · Downloads, stars and installs are aggregated monthly from public skill registries (ClawHub, SkillHub). See methodology.