SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsray-cover

ray-cover

Design & creation

View repository

Updated 10-15 22:37

About this skill

Ray Cover

Treat the cover as a visual compression of the article's judgment, not a decorative image. First confirm what the article is saying, then find a metaphor; by default let Image 2 handle both the imagery and the Chinese text, and generate separately for WeChat Official Account, regular X, and X Article, without cropping or stretching from the same finished piece.

Fixed Principles

  1. A cover expresses only one judgment or relationship, and keeps only one main metaphor.
  2. Do not use generic AI icons such as "AI brain, glowing chip, robot, neon gradient, floating UI" in place of the point of view.
  3. Do not imitate the recognizable signature style of a specific brand or living artist. Write "Vox" uniformly as "editorial metaphor collage," and break the Adrian reference down into reusable composition, color, and material principles.
  4. By default, use Image 2 of the system's image generation capability to directly generate the complete cover. Call it once separately for each platform, specifying the target size, ratio, and safe area in the prompt.
  5. After generation, check the metaphor, composition, title position, edge safe area, and thumbnail readability. Chinese font, stroke, or occasional glyph differences are not grounds for rejection, retry, or fallback.
  6. scripts/compose_cover.py no longer serves as an automatic fallback. Use it only when the user explicitly requests post-production typesetting, pixel-level cross-platform consistency, or has already provided a text-free base image.
  7. The Chinese main title itself is the main visual. Whether generated by Image 2 or typeset in post-production, it must establish hierarchy through font size, typeface, color, and position, and must not be shrunk into an image caption.
  8. WeChat Official Account, regular X share images, and X Article 5:2 share one visual motif, but are generated separately for their respective sizes; do not stretch, crop, or reuse another platform's finished piece.
  9. Do not auto-publish, do not overwrite existing covers; when a file with the same name exists, use a new version number.

Workflow

1. Read the article basis

Read at least the complete body text; if a finished draft exists, also read the one-sentence judgment, emotional core, core conflict sentence, and dissemination sentence. When the body text's judgment is not yet stable, first return to ray-writer, and do not use the cover to make decisions for the article.

Extract:

  • One-sentence judgment
  • The emotion readers feel upon entering
  • The conflict most worth being seen
  • A cover short title of 8–18 Chinese characters
  • A subject-predicate relationship that can be drawn

2. Choose a visual direction

Read style-system.md. Exploration mode gives three clearly distinct directions; production mode automatically selects an approved direction based on the article structure. Stop to confirm only when the visual judgment is unstable, involves a real person's likeness, or the user explicitly wants to choose.

Default selection rules:

  • Abstract viewpoints, trust, identity, responsibility:editorial-metaphor-collage or minimal-metaphor
  • Methodology, systemic judgment:midcentury-editorial
  • Restrained, anti-AI noise, strongly text-oriented:quiet-ink
  • Connections between two products, tools, or processes:tool-bridge
  • Material experiments are used only when the user explicitly requests them

3. Write the cover task package

Build JSON according to cover-brief.md . By default write generation_strategy: direct-first, the cover copy, the shared visual motif, and the complete prompt for each platform in platform_prompts . Run:

python3 scripts/brief_check.py <cover-brief.json>

If it fails, fix it; do not skip it. The task package must retain the article path, title, judgment, metaphor, style, generation strategy, cover copy, prompts for each platform, prohibited items, and platform outputs. When entering X Articles,outputs and platform_prompts must both explicitly declare the x_article 5:2 finished product; it cannot merely be noted afterward in the manifest.

4. Generate the complete cover directly for each platform separately

For the WeChat Official Account, regular X, and X Article, call the system image generation capability Image 2 once each. Each prompt must clearly specify:

  • Purpose, target pixel dimensions, and platform aspect ratio
  • One main metaphor, visual position, and text safe area
  • Section, main title line breaks, and subtitle
  • Font character, font size hierarchy, color, and relative position
  • Do not add other text
  • Prohibit brand logos, watermarks, garbled text, and irrelevant elements

Each platform is generated independently from the same visual motif; do not crop one platform's original image into another platform's. Open the original image to check the subject and safe area, then scale it down to the platform preview size to check two-second readability. Retry for that platform only when the metaphor does not hold, the composition is cropped, or the title clearly leaves the safe area; do not retry because of Chinese font or glyph details.

If the pixel dimensions returned by Image 2 differ slightly from the target, only scale proportionally to the target size at the same aspect ratio; if the aspect ratio does not match, regenerate for that platform, and do not crop across platforms.

5. Use deterministic typesetting only when the user explicitly requests it

Deterministic typesetting only handles post-production typesetting tasks explicitly chosen by the user, and is not automatically triggered by Chinese font details from Image 2. When using it, first read chinese-typography.md, then output separately by platform:

python3 scripts/compose_cover.py --background <底图> --title "主标题上半|主标题下半" --subtitle "副标题" --eyebrow "栏目" --platform wechat --layout left --out <公众号封面.png>
python3 scripts/compose_cover.py --background <底图> --title "主标题上半|主标题下半" --subtitle "副标题" --eyebrow "栏目" --platform x --layout left --out <X封面.png>
python3 scripts/compose_cover.py --background <底图> --title "主标题上半|主标题下半" --subtitle "副标题" --eyebrow "栏目" --platform x-article --layout left --out <X Article封面.png>

When the Chinese title has a clear core word, use --title-mode keyword --keyword <核心词>; when there is no core word, split it into two lines by semantics and use | in the title to control line breaks.

6. Generate versions for each platform and check

Read platform-output.md. In direct generation mode, specify the pixel dimensions, aspect ratio, safe area, and text hierarchy separately for each platform; keep the visual motif consistent, but use an independent prompt and independent generation result for each platform. Do not require uniformity when there are reasonable differences in composition or fonts; enter deterministic typesetting only when the user explicitly requests pixel-level consistency.

Each finished product must be checked:

  • The title, subtitle, and section are within the safe area, with no extra text interference
  • The file dimensions are accurate, and the text and main visual are not cropped
  • The main title is still clear in a thumbnail 400–420px wide
  • The metaphor still holds at each aspect ratio
  • It does not contain logos not provided, implications of real people, or generic AI clichés

Only when the user requests sending the X Article cover to the backend does it enter the editor Preview check; when only producing local assets, do not open the backend.

7. Save to disk and hand off

By default, save to:

<vault>/60-素材/10-图片/10-封面/<YYYY-MM-DD-主题>/

<vault> comes from the user's explicit path,.ray-obsidian.json or a compatible knowledge base structure. If there is no knowledge base, save to the user-specified current workspace, and do not create a vault with a fixed name on your own.

Direct generation mode retains:

  • cover-brief.json
  • cover-brief.md
  • direct-source-<platform>.png(the model's original image generated independently for that platform)
  • wechat-cover.png
  • x-cover.png
  • x-article-cover.png(when it needs to be written to the X Articles backend)
  • Thumbnails for each platform
  • manifest.json

Post-production typesetting mode additionally retains background-<platform>.png, and records in the manifest why the user chose post-production typesetting.

If the user also needs about 5 seconds of vertical dynamic material, distill a wordless visual description from the same metaphor and hand it off to the already installed gbro-collage-broll; respect its three confirmations and do not bypass them within this Skill. Only consider a process like vox-director for a complete narrated video.

Delivery requirements

Deliver the required platform covers, task packages, and the final prompts for each platform. Explain that each platform is generated separately by Image 2, or whether the user explicitly chose post-production typesetting; describe the metaphor and visual direction in one sentence, and indicate whether the dynamic material still needs to be produced.