LiamGvchi /

gc-minimal-zine-poster

编程 Skill 图片与设计 Skill

趋势观察中

Codex skill for generating quiet minimal zine-style editorial poster prompts and images.

原始仓库简介 · 中文摘要待补充 ·

GitHub 仓库 ↗

README / 项目原文

README.md ↗抓取于 2026/10/24 00:49:26

GC Minimal Zine Poster v0.3.1

English · 简体中文 · 日本語

Minimal Zine Poster v0.3.1 is a Codex skill for turning a theme, sentence, article idea, object, mood, photograph, or reference set into a sparse paper-texture editorial poster, a production-ready image prompt, or a reusable visual system.

The callable skill name is gc-minimal-zine-poster-v0-3.

What's New in v0.3.1

Replaces ambiguous composition wording with concrete focal-element and color-accent language.
Requires the final image prompt to name the visible carrier, such as a photo crop, paper cutout, ink block, specimen, printed illustration, texture window, or typography.
Adds quality checks for unrequested maritime symbols and generic pictograms.
Adds a regression eval for a non-maritime theme with a bright red focal element.
Keeps Generate, Reference Analysis, Prompt-only, Analyze + Generate, and Photo Input routes unchanged.
Keeps the callable Skill name gc-minimal-zine-poster-v0-3 for backward compatibility.

Visual Direction

The skill compiles each request into a sparse vertical paper poster with:

a default 3:5 aged-paper canvas
70%-90% negative space
one small imageable subject or visual event
serif, typewriter, monospaced, or restrained small sans-serif typography
one clearly visible high-chroma color accent
xerox, risograph, halftone, letterpress, or scanned-paper defects
a quiet Japanese/Korean indie-zine or minimal editorial mood

It avoids commercial advertising layouts, glossy mockups, cinematic lighting, 3D rendering, neon, dense scrapbooks, copied reference identity, and long clean text blocks.

Examples

All six examples below were created by the repository author.

Night Door	Yellow Step

Shore Pause	Pause Map

Typhoon Memory	Moon Tide

Requirements

Codex or another compatible Skill runtime.
Image inspection for reference analysis and quality review.
Built-in image generation for Generate and Analyze + Generate modes.

The Skill package contains no scripts, external fonts, API keys, private paths, or downloaded runtime assets. Image-generation quality and photo preservation still depend on the model available in the host environment.

Installation

Clone the current v0.3.1 release into a matching Skill directory:

git clone https://github.com/LiamGvchi/gc-minimal-zine-poster.git \
  ~/.codex/skills/gc-minimal-zine-poster-v0-3

Restart Codex if the Skill does not appear immediately.

Upgrading from v0.3.0

The callable Skill name and installation directory remain unchanged. If your existing v0.3 installation is a clean Git checkout, update it with:

git -C ~/.codex/skills/gc-minimal-zine-poster-v0-3 pull --ff-only

If the directory is not a Git checkout or contains local edits, install a fresh copy separately and compare your changes before replacing anything.

Upgrading from v0.1

Do not pull v0.3.1 into an existing ~/.codex/skills/gc-minimal-zine-poster-v0-1 directory. The folder name and the Skill frontmatter name must stay aligned.

Install v0.3.1 beside the old copy using the command above. After confirming v0.3.1 works, you may keep or remove the old directory yourself.

To install the preserved v0.1 release:

git clone --branch v0.1.0 \
  https://github.com/LiamGvchi/gc-minimal-zine-poster.git \
  ~/.codex/skills/gc-minimal-zine-poster-v0-1

Usage

Generate a poster:

Use $gc-minimal-zine-poster-v0-3 to make a poster about a rainy secondhand bookstore.

Analyze references without generating:

Use $gc-minimal-zine-poster-v0-3 to analyze this image folder, separate fixed and variable rules, and return a reusable prompt without copying the source text.

Use a supplied photograph as an edit target:

Use $gc-minimal-zine-poster-v0-3 to turn this portrait into a poster. Preserve the person's identity and clothing; change only the layout and paper treatment.

Request prompt-only output:

Use $gc-minimal-zine-poster-v0-3 to return only the final image-generation prompt for an old bookstore closing at night.

Request Modes

Generate: content → visual metaphor → prompt → generated raster → inspection.
Photo Input: assign image roles and preservation levels, pass the actual target image into generation, and inspect source preservation.
Reference Analysis: inspect real files and return observed evidence, fixed rules, variable rules, sample residue, a reusable prompt, and limitations.
Prompt-only: return a final four-paragraph prompt and recipe without claiming an image was generated.
Analyze + Generate: extract a visual system first, then generate a new composition without copying a source layout.

Output

Generation requests return:

the generated raster poster
the final image-generation prompt
the selected recipe
a short interpretation note
photo role and preservation details when a photograph is supplied

Reference-analysis requests return the evidence-based visual system and reusable prompt without generating an image unless generation was also requested.

Repository Structure

SKILL.md: request routing and execution workflow
references/style-system.md: fixed identity, variables, color, subject logic, and anti-identity
references/prompt-compiler.md: prompt field order and four-paragraph compiler
references/variation-engine.md: layout and variation recipes
references/reference-analysis.md: evidence and synthesis protocol
references/quality-gate.md: generation, reference-analysis, and prompt-only checks
agents/openai.yaml: Codex UI metadata
evals/evals.json: reusable evaluation prompts
examples/: six author-created poster examples
LICENSE: MIT license

Acknowledgements

Thanks to Xiaohongshu user @李李 for reporting that an abstract composition term could be interpreted literally and introduce an unintended nautical icon. This feedback led directly to the v0.3.1 terminology, prompt-compiler, quality-gate, and regression-eval fixes.

License

MIT. See LICENSE.

README.zh-CN.md ↗抓取于 2026/10/24 00:49:26

GC Minimal Zine Poster v0.3.1

English · 简体中文 · 日本語

Minimal Zine Poster v0.3.1 是一个 Codex Skill：它可以把主题、句子、文章构想、物件、情绪、照片或参考图集，转化为一张纸张质感的极简编辑海报、一条可直接使用的生图 Prompt，或一套可复用的视觉系统。

调用名称为 gc-minimal-zine-poster-v0-3。

v0.3.1 更新内容

用明确的视觉重心与色彩强调语言，替代可能被按字面误解的抽象构图术语。
要求最终生图 Prompt 写清实际可见载体，例如照片裁切、纸片、墨块、标本、印刷插图、纹理窗口或字体。
质量检查新增无关船用符号与通用图标检查。
新增一条非海事主题、鲜红色视觉重心的回归评测。
保留 Generate、Reference Analysis、Prompt-only、Analyze + Generate 和 Photo Input 路线。
为了向后兼容，调用名称继续使用 gc-minimal-zine-poster-v0-3。

视觉方向

这个 Skill 会把请求编排成一张留白充足的竖版纸张海报，主要特征包括：

默认 3:5 比例的仿旧纸张画布
70%–90% 的留白
一个小型、可被清楚表现的主体或视觉事件
衬线、打字机、等宽或克制的小号无衬线字体
一个清晰可见的高饱和度色彩强调区域
复印、孔版印刷、网点、凸版印刷或扫描纸张的瑕疵与质感
安静的日式／韩式独立 ZINE 或极简编辑设计氛围

它会避开商业广告式布局、光亮样机、电影感布光、3D 渲染、霓虹、密集拼贴、复制参考图身份以及大段整齐文字。

示例

以下六张示例图均由仓库作者制作。

Night Door	Yellow Step

Shore Pause	Pause Map

Typhoon Memory	Moon Tide

使用要求

Codex 或兼容的 Skill 运行环境。
参考图分析和质量检查需要图片读取能力。
生成模式与分析后生成模式需要运行环境提供图片生成能力。

Skill 包中不包含脚本、外部字体、API 密钥、私有路径或需要下载的运行资源。最终生图质量和照片保真程度仍取决于运行环境中的图片模型。

安装

把当前 v0.3.1 克隆到名称一致的 Skill 目录：

git clone https://github.com/LiamGvchi/gc-minimal-zine-poster.git \
  ~/.codex/skills/gc-minimal-zine-poster-v0-3

如果 Skill 没有立即出现，请重启 Codex。

从 v0.3.0 升级

调用名称和安装目录保持不变。如果现有 v0.3 安装是干净的 Git 仓库，可以执行：

git -C ~/.codex/skills/gc-minimal-zine-poster-v0-3 pull --ff-only

如果该目录不是 Git 仓库，或包含本地修改，请先安装一份全新副本并比较差异，不要直接覆盖自己的改动。

从 v0.1 升级

不要在已有的 ~/.codex/skills/gc-minimal-zine-poster-v0-1 目录里直接拉取 v0.3.1。文件夹名必须与 Skill frontmatter 中的名称保持一致。

请使用上面的安装命令，把 v0.3.1 安装在旧版本旁边。确认 v0.3.1 可以正常调用后，你可以自行决定保留或删除旧目录。

如果仍然需要安装保留的 v0.1：

git clone --branch v0.1.0 \
  https://github.com/LiamGvchi/gc-minimal-zine-poster.git \
  ~/.codex/skills/gc-minimal-zine-poster-v0-1

使用方法

生成海报：

用 $gc-minimal-zine-poster-v0-3 做一张关于雨天旧书店的海报。

只分析参考图，不生成图片：

用 $gc-minimal-zine-poster-v0-3 分析这个图片文件夹，区分固定规则和可变规则，不要复制原图文字，并给我一条可复用 Prompt。

把用户照片作为编辑目标：

用 $gc-minimal-zine-poster-v0-3 把这张人物照片做成海报，保留人物身份和服装，只改变排版与纸张质感。

只要最终 Prompt：

用 $gc-minimal-zine-poster-v0-3 写一条关于旧书店夜晚关门的最终生图 Prompt，不要出图。

请求模式

**Generate：**内容 → 视觉隐喻 → Prompt → 位图生成 → 结果检查。
**Photo Input：**确定图片角色和保真等级，把真实目标图片传入生成，并检查主体保留情况。
**Reference Analysis：**检查真实文件，返回观察证据、固定规则、可变规则、样本残留、可复用 Prompt 和限制。
**Prompt-only：**只返回四段式最终 Prompt 与 Recipe，不声称已经生成图片。
**Analyze + Generate：**先提取视觉系统，再生成不复制原图构图的新作品。

输出

生成请求会返回：

生成的位图海报
最终生图 Prompt
所选 Recipe
一段简短的内容解释
提供照片时的图片角色和保真信息

参考图分析请求会返回基于证据的视觉系统和可复用 Prompt；除非用户同时要求生成，否则不会出图。

仓库结构

SKILL.md：请求路由与执行工作流
references/style-system.md：固定风格、可变项、色彩、主体逻辑和避免方向
references/prompt-compiler.md：Prompt 字段顺序和四段式编译器
references/variation-engine.md：布局与变化方案
references/reference-analysis.md：参考图证据与综合协议
references/quality-gate.md：生成、参考分析和仅 Prompt 的质量检查
agents/openai.yaml：Codex 界面元数据
evals/evals.json：可复用评测 Prompt
examples/：六张由作者制作的海报示例
LICENSE：MIT 许可证

致谢

感谢小红书用户 @李李 指出：提示词中的抽象构图术语可能被模型按字面理解，生成无关的船用图标。这条反馈直接促成了 v0.3.1 对术语、Prompt 编译器、质量检查和回归评测的系统修复。

许可证

MIT。详见 LICENSE。