lgboom-xhs-cover-creator

star 0

Use when the user wants to turn an article, post draft, topic idea, or Chinese long-form content into a high-quality Xiaohongshu cover illustration. Covers English prompt writing, the user's preferred quiet hand-drawn illustration style, 3:4 cover composition, 1242x1660 export expectations, and PackyAPI gpt-image-2 / image-2 generation through the OpenAI Images API. Trigger for: 小红书封面, 插画封面, 文章配图, 根据文章生成图, image-2 绘图, PackyAPI 生图, GPT-Image-2, 3:4 封面图, 1242x1660.

Himog0921 By Himog0921 schedule Updated 6/2/2026

name: lgboom-xhs-cover-creator description: > Use when the user wants to turn an article, post draft, topic idea, or Chinese long-form content into a high-quality Xiaohongshu cover illustration. Covers English prompt writing, the user's preferred quiet hand-drawn illustration style, 3:4 cover composition, 1242x1660 export expectations, and PackyAPI gpt-image-2 / image-2 generation through the OpenAI Images API. Trigger for: 小红书封面, 插画封面, 文章配图, 根据文章生成图, image-2 绘图, PackyAPI 生图, GPT-Image-2, 3:4 封面图, 1242x1660.

小红书封面插画生成器

把用户文章变成一张可用于小红书封面的插画图。核心要求是:先理解文章,再设计视觉隐喻,最后用高质量英文提示词调用 PackyAPI 的 gpt-image-2 出图。

工作流

  1. 读取文章,提炼一个核心情绪和一个核心冲突。

    • 不要把整篇文章塞进画面。
    • 只保留一个能被一眼看懂的视觉隐喻。
  2. 设计封面画面。

    • 默认不在图片里放字,除非用户明确要求。
    • 给后期标题留下干净留白,优先上方或右上方。
    • 尺寸目标是小红书封面 1242x1660,生成阶段优先用 1536x2048,最后导出为 1242x1660
  3. 写英文提示词。

    • 使用完整英文,不要中英混杂。
    • 必须包含:主体、动作关系、象征物、画面情绪、构图比例、材质和排除项。
    • 使用 references/illustration-style.md 的风格词,不要临时换成商业矢量、动漫、3D 或摄影风格。
  4. 调用图片模型。

    • 优先使用 scripts/generate_packy_image.py
    • 需要 API Key 时从环境变量读取,优先 PACKY_API_KEY,也支持 PACKY_SORA_API_KEY
    • 没有密钥时,只输出英文提示词和可执行命令,不要声称已经生成图片。
  5. 交付结果。

    • 如果成功生成:给用户最终图片路径、原始图路径、提示词路径,并说明用户现在应该看到什么。
    • 如果失败:用非技术语言说明出了什么问题、影响是什么、下一步怎么处理。

英文提示词结构

按这个结构写,不要输出空泛审美词堆砌:

Create a vertical 3:4 editorial illustration for a Xiaohongshu cover.

Scene: [one concrete symbolic scene based on the article].
Characters and actions: [who appears, what they are doing, how their relationship is shown visually].
Metaphor: [what the image is really saying, without writing text in the image].
Composition: generous negative space for a later title overlay, quiet off-white paper background, main figures placed [left/right/lower third/etc.].
Style: hand-drawn contemporary book illustration, thin uneven ink outlines, muted colored-pencil fills, visible paper grain, subtle cross-hatching, flat perspective, restrained colors.
Mood: [tender / uneasy / intimate / restrained / quietly surreal].
Do not include: text, captions, logos, watermarks, glossy vector art, anime style, 3D render, photorealism, busy background, perfect corporate illustration.

PackyAPI 调用

先读 references/packy-gpt-image-2.md。最常用命令:

PACKY_API_KEY="$PACKY_API_KEY" \
python3 ".claude/skills/lgboom-xhs-cover-creator/scripts/generate_packy_image.py" \
  --prompt-file "/absolute/path/to/prompt.txt" \
  --out-dir "/absolute/path/to/output-dir"

无密钥或只想检查请求时:

python3 ".claude/skills/lgboom-xhs-cover-creator/scripts/generate_packy_image.py" \
  --prompt "Create a vertical 3:4 editorial illustration..." \
  --dry-run

质量门禁

出图后检查四件事:

  • 画面是否只讲一个视觉隐喻。
  • 插画是否有纸张肌理、细线条、低饱和色块和手绘感。
  • 是否留出了标题空间。
  • 是否已经导出 1242x1660,没有拉伸变形。

不合格时,优先改英文提示词后重试;不要用中文短提示词反复抽卡。

Install via CLI
npx skills add https://github.com/Himog0921/topic-dashboard --skill lgboom-xhs-cover-creator
Repository Details
star Stars 0
call_split Forks 0
navigation Branch main
article Path SKILL.md
More from Creator