メニュー
Analyze an uploaded or local reference poster, extract its reusable design system, and transpose that system to an original poster with a new theme, title, subtitle, or campaign copy.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/jiemianduan/poster-style-transfer --skill poster-style-transferコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SOC 職業分類に基づく
SKILL.md を表示中
| name | poster-style-transfer |
| description | Analyze an uploaded or local reference poster, extract its reusable design system, and transpose that system to an original poster with a new theme, title, subtitle, or campaign copy. |
| license | MIT |
| compatibility | Works with Codex and Agent Skills-compatible runtimes. Direct generation requires image viewing and image-generation capabilities. |
| metadata | {"author":"jiemianduan","version":"0.1.0"} |
Extract poster grammar, not poster content. Preserve the reference's design system while rebuilding its subject matter, copy, symbols, and narrative for the user's new theme.
Infer the mode from the request:
Poster Style Spec, not a general aesthetic review.If the user invokes the skill but only asks “what style is this?”, explain that the skill produces a reusable production specification and return the compact Style Spec. Do not turn the response into art-history commentary.
Require at least one usable poster image. For a local image, inspect it at the highest available detail before analysis.
Confirm that the reference has poster characteristics such as a deliberate canvas, typography hierarchy, hero visual, information modules, and designed reading order. If it is not a poster, ask for a poster reference instead of applying the workflow to a generic image.
Treat one image as the primary reference by default. If several posters are supplied, identify which one controls layout, typography, hero treatment, and color; do not silently average contradictions.
Require only:
Accept these optional fields: main title, subtitle, supporting copy, date, location, target audience, poster purpose, aspect ratio, brand assets, and target image model.
Infer tasteful missing details when the user authorizes direct generation. Do not invent factual addresses, prices, schedules, official slogans, logos, or endorsements. Label invented campaign information as concept copy when it could be mistaken for a real announcement.
Read poster-style-spec.md and fill every high-impact field supported by visible evidence. Classify the reference using poster-archetypes.md, then prioritize the dimensions associated with that archetype.
Analyze in this order:
Use relative measurements such as percentages of canvas width or height. Prefer “title occupies the upper-left 38%” over “large title at the top.” Describe visible behavior rather than guessing font names, artists, brands, lenses, engines, or original generation settings.
Create three lists before writing the generation prompt.
Preserve the reference's highest-impact design rules: aspect ratio, grid family, margin rhythm, title footprint, hero-to-canvas ratio, dominant overlap behavior, palette relationship, typography category, primary image treatment, and density map.
Preserve with adaptation: decoration count, secondary modules, texture intensity, lighting direction, local rotations, small labels, and accent shapes.
Replace the reference's people, characters, products, props, scenery, logos, wording, brand identifiers, copyrighted character designs, and theme-specific symbols. Do not trace a unique illustration or reproduce the complete composition pixel for pixel.
Preserve spatial grammar rather than exact coordinates. Adapt the grid when the new title length, script, or hero silhouette requires it.
Convert the user's theme into a coherent content system before prompting:
Choose elements that belong to the new theme and can be rendered through the reference's visual technique. Keep every prop purposeful. Do not carry over a reference object merely because it is visually prominent.
Write a structured prompt in this order:
output intent → reference role → new theme → locked layout → typography → new hero → environment/props → color → material/lighting → exact text → constraints/avoid list
State that the input image is a style and design-system reference only. Explicitly list both the traits to preserve and the content to replace.
Quote all literal copy. Require verbatim rendering and prohibit extra text. Minimize in-image copy when the user has not supplied it.
For short headlines, render text in the image. For dense copy or exact Chinese typography, prefer a two-layer workflow: generate the visual with reserved text zones, then add exact copy using a deterministic vector, HTML, or document-layout tool when available. Do not claim text accuracy without checking the image.
Use the available image-generation tool for direct generation. Pass the reference as a style reference, not an edit target. Never request preservation of the reference's specific person, logo, wording, or proprietary character.
After generation, inspect:
Read fidelity-rubric.md and score the result. If a single issue dominates, make one targeted revision and repeat the invariants. Do not rewrite the entire prompt for a local text or placement defect.
If the built-in image service fails, retry once. Do not switch to a CLI/API path that requires credentials unless the user explicitly authorizes it. If both the reference path and pure-generation path fail, return the final prompt and report the service failure clearly.
Return:
Return:
Poster Style Spec;Do not generate an image.
Return:
Include at least: [主题], [英文主标题], [中文主标题], [副标题], [主视觉], [动作], [主题道具], [主色], [强调色], [日期], [地点], [宣传语], and [画幅比例].