Design & Motion · dacnay816y62-hub/photo-revival

photo-abstract-editoria

Name: photo-abstract-editoria
Author: dacnay816y62-hub/photo-revival

Turn ordinary photos or everyday snapshots into poetic hand-drawn illustrations with large white-paper negative space, vivid localized color, subtle collage accents, and tiny handwritten captions. Use when a user wants a real photo reimagined as an illustration rather than filtered, beautified, or photorealistically edited.

L1 Basic · Source registration onlyV1 Commit-PinnedDual-pass Audit

Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

It is a pure rule/specification-based skill: SKILL.md consists entirely of style constraints, workflows, and prompt structures, containing no scripts, executable code, or tool calls—the actual "image generation" step is explicitly delegated to image tools in the host environment, while this repository only handles organizing design judgments.

README.md

**生成说明：** Skill 组织设计判断、提示词与执行流程；图片由当前环境中可用的图像工具生成或编辑。示例用于理解视觉方向，具体来源以本仓库记录为准，不能据此保证每次得到相同效果。

What it is doing: The author intentionally decouples the "design specification" from the "image engine"—the repository self-describes as not including image services or generation quotas (README notes "Installation does not include image services, accounts, or generation quotas; actual image output depends on the environment you use"), and the "Output prompt shape" section in SKILL.md only states "When directing image generation, describe: …", meaning it directs existing text-to-image/image-to-image tools in the host.

The core of the effect is a set of quantifiable, verifiable hard layout constraints: 3:4 vertical composition + subject occupying only 10-16% of the page (absolute maximum 18%) + 80-88% negative space + localized color, turning "poetic negative space" from an adjective into numerical discipline with percentages.

SKILL.md

Default to: - 3:4 vertical composition - large white-paper negative space - a tiny subject or focal cluster occupying about 10-16% of the page - delicate hand-drawn texture - vivid local color only where it matters - tiny handwritten text, dates, or short captions

Step 4 of the Workflow provides the paired constraint "Leave 80-88% of the page as blank white paper."; Style rules further restrict "Keep the strongest color inside that small illustrated area; do not let color spread into the blank field."—binding values to space allows the generated results to be visually inspected and accepted.

The negative list (Avoid) and deviation self-correction take on the primary corrective role: the most common runaway states in AI image generation (full-bleed fills, photo filter looks, heavy poster typography, subjects exceeding 18%, flat vectors, large collages) are prohibited line-by-line, while the README FAQ provides stricter English regeneration phrasing targeting "subject too large / scrambled text / looks like a photo filter / color spread too full".

README.md

### 主体太大怎么办？重新生成，并把约束写得更硬： ```text The illustrated subject must occupy only 10-16% of the full page. Absolute maximum subject area is 18%. Keep 80-88% of the page blank white paper. ```

The original text of the Avoid section in SKILL.md reads: "Full-bleed image fills / Photo-filter looks / Heavy poster typography / Dense layouts / Subject areas larger than 18% of the page / Flat vector treatment / Large collage blocks / Overly realistic retouching"—the same values appear repeatedly across rules and correction sections, creating a "generate → inspect → regenerate" closed loop.

Text is strictly limited to "tiny handwritten captions" rather than poster titles: one short Chinese poetic line + one FIELD NOTE/DATE-style small English text, and the README explicitly warns that text must be small, minimal, handwritten-like, and must not serve as a headline—this is the boundary distinguishing it from illustration posters or graphic design skills.

README.md

### 文字重复或乱码怎么办？减少文字，只保留一句中文和一个英文日期。文字必须“小、少、像手写批注”，不要当成海报标题。

Step 5 of the SKILL.md Workflow applies a matching constraint: "Add text only as a small note, label, date, or poetic fragment."; the README also includes a built-in bilingual micro-copy library (Chinese poetic lines like "圆镜里藏着今天" [Today hides in the round mirror], English archival short phrases like FIELD NOTE / 08.02) for the model to pick from, reducing the probability of invented garbled text.

All operational layers are consolidated in the README: installation (including Codex user/project-level skill directory conventions and the requirement that "the folder must be named photo-revival"), a seven-part general English prompt template (Preserve/Redraw/Composition/Texture/Color/Text/Avoid), and batch testing with quality check workflows—forming a two-file structure of "SKILL.md = rules, README.md = Chinese operation manual".

README.md

仓库名与调用名可能不同，以上以 `SKILL.md` 中的名称为准。安装不包含图像服务、账户或生成额度；实际出图取决于你使用的环境。

The ready-to-use seven-part template provided in the README translates each rule from SKILL.md into English instructions that can be directly pasted to the image model; steps 6–7 of the recommended workflow require "Check if the subject is too large. If it exceeds 18%, regenerate" and "Check if the text is oversized, repetitive, or garbled. If there is an issue, regenerate or reduce text"—embedding quality gates directly into the process rather than relying on model self-discipline.

2Core Capabilities

01Actionable style translation rules for converting photos into small hand-drawn illustrations on white paper (subject identification → simplified retention → paper texture/brushstroke reconstruction → negative space composition)

02Quantified layout discipline: dual numerical constraints of 10-16% subject coverage (18% upper limit) and 80-88% negative space

03Localized color control: intense colors are only allowed within the small illustration area and forbidden from bleeding into the blank paper

04Tiny handwritten text caption specifications (short Chinese/English phrases or dates, minimal, handwriting-like, poster headlines prohibited)

05Ready-to-use seven-part English image prompt template (Preserve/Redraw/Composition/Texture/Color/Text/Avoid)

06Deviation self-correction toolkit: hard-constraint regeneration prompts for four common failures (subject too large, garbled text, filter-like appearance, oversaturated color spread)

07Built-in Chinese poetic caption library and English FIELD NOTE archival phrase library (ready for direct selection)

08Batch testing and manual quality assurance workflow (including an acceptance gate for "regenerate if subject > 18%")

4Recommended Governance & Evidence

Suggested Controls for Safe Operation

For Agent Runtime

Can be auto-loaded

For Administrators

Evidence & Triggers (File Paths & Excerpts)

SKILL.md: Pure prompt instructions and rule text, without script execution or external impacts

Audit Notes: . Pure prompt-rule skill: the git tracking manifest contains only 21 files (LICENSE, README.md, SKILL.md, agents/openai.yaml, plus 17 sample PNGs). The entire repository involves no script execution, no outbound network requests, no credential reading, and no local file writing; the only URLs are concentrated in README documentation links (GitHub ZIP download and OpenAI official installation docs), with no runtime invocation endpoints. Image generation is delegated to host image tools, whose outbound data boundaries are governed by the host environment and do not constitute risks introduced by this repository.

5Second-Pass Verification

[ok] 文件资产面（是否有隐藏脚本/混淆内容） — 第二遍 git ls-files 复核：21 个跟踪文件 = LICENSE、README.md、SKILL.md、agents/openai.yaml、examples/ 下 17 张 PNG，与第一遍目录扫描一致；无任何脚本、无隐藏文件、无子模块。
[ok] 网络/执行/凭证 token 扫描 — 全仓正则扫描 curl|wget|http|bash|exec|subprocess|api_key|secret|token|.env|credential|ffmpeg|python|openai|PIL|write( 等：命中仅 README.md 内 https:// 文档链接 5 处（GitHub 仓库列表、ZIP 下载、learn.chatgpt.com 官方安装文档、姊妹 skill 链接）+ markdown 语法；SKILL.md、agents/openai.yaml 零命中。
[ok] 引文真实性（逐条字节级回源） — 18 条拟用 quote 全部通过 python 字节包含校验（OK 18/18），含多行 quote（Workflow 编号步骤、README FAQ 代码块）；无 paraphrase 冒充。
[ok] 功能声明 vs 实际能力夸大检查 — 无夸大：skill 自称只『组织设计判断、提示词与执行流程』，出图工具、图像服务与额度均不在仓库内；唯一注意点——SKILL.md 标题『Photo Revival』与 description 用第一人称式能力描述（Turn ordinary photos into…），若脱离 README 单独阅读可能被误认为自带出图能力，实际执行权在宿主图像工具。
[ok] 元数据一致性（name/license/commit） — SKILL.md frontmatter name = photo-revival = 仓库名 photo-revival（本地镜像目录名 photorevival 为缩写）；LICENSE 本地文件 MIT（Copyright (c) 2026 FANTASY）与 GitHub API spdx MIT 一致；git HEAD == 任务 pin 5574fa2f6fa8fa52b232d7836b7960bf6b5753f6 == GitHub main pushed_at 2026-09-05T21:33:07Z。注：本分析入库展示名 photo-abstract-editoria 与仓库/frontmatter 名 photo-revival 不一致（任务给定展示名），JSON 内容仍以真实源码为准，文件名按任务名入库。

6Verdict & Scenarios

All constraints are quantified and verifiable: 10-16% subject size (max 18%) and 80-88% negative space make it possible to objectively assess whether deviations have occurred

Comprehensive correction system: the Avoid negative list and FAQ address four common failure modes with "stricter" English regeneration phrasings, turning quality gates into actionable steps

Honest operational boundaries: clearly acknowledges that images are generated by host image tools and that the repository includes neither image services nor quotas, without exaggerating its capabilities

High out-of-the-box readiness: the README provides a bilingual copy library, a seven-part prompt template, installation guidance, and 17 sample images, allowing Chinese users to follow along directly

Best for: Ideal for agent hosts with image generation capabilities (such as Codex or ChatGPT variants) to transform casual snaps, discarded shots, and everyday life photos into artistic illustration pages featuring "large white paper negative space + small hand-drawn artwork + poetic micro-captions"; also suitable for skill authors who want to reuse the authoring technique of "quantified layout constraints + Avoid lists + deviation regeneration FAQ" in their own visual skills. The Chinese prompts and copy library are particularly friendly to Chinese users.

Not for: Not suitable for hosts without image generation capabilities; not suitable for portrait retouching, commercial posters, dense infographics, full-bleed collages, or restoration tasks requiring pixel-level photo fidelity (explicitly listed in the README under "What it is not suitable for"); not suitable for automated pipelines aiming for "one-shot generation with zero human inspection"—it requires manual verification of layout proportions and text.

Install Agent-Ready

① Verified Mirror 2026-09-09

Option A · Direct DownloadDownload photo-abstract-editoria.tar.gz

sha256: 18ee29cf2ca902fd…

Option B · Copy Guide for Agent

install.json

Agents automatically download and verify mirrors from this JSON guide.

② Upstream GitHub

GitHub Repo ↗

Pinned to commit 5574fa2f6f.