Use baoyu-cover-image for a single visual representing an article. Its selectable dimensions make the brief more precise than asking for something attractive. Supply the publication context and title so composition can serve the actual use.
Install and invoke
Run this in your project terminal to install only this skill for Claude Code:
npx skills add jimliu/baoyu-skills --skill baoyu-cover-image -a claude-code
Node.js and npx must be available. For Codex, replace -a claude-code with -a codex. In the agent conversation, explicitly ask it to use baoyu-cover-image and supply your material. The installation command belongs in the terminal, not in your content brief.
Provide the title and display context
Supply the article, exact title, intended ratio, and purpose of any references. A wide listing thumbnail, square social card, and large article header have different needs. Choose the primary use first.
Example brief:
Use baoyu-cover-image for a 16:9 cover of article.md.
Keep the original title without a subtitle. Choose conceptual type, cool palette, flat-vector rendering, title-only text, and subtle mood.
Confirm the plan and available backend first.
Show a team bringing scattered feedback into one place, without filling the image with software logos.
Options include --type, --palette, --rendering, --text, --mood, and --aspect. For a visual without lettering, use --text none or --no-title; the page can still display its normal heading. First-run preferences are saved in the skill's EXTEND.md. Check that colors or watermarks inherited from another project are appropriate.
Verify backend, references, and output
Rendering needs an available image backend. If none is provided natively, configure a separate baoyu-image-gen installation. Its Google path requires an accessible model and your API credential through GOOGLE_API_KEY. Verify one image before planning larger production.
Explain whether references supply palette, composition, or direct image input. Direct input requires backend support; a written description does not mean the original image was transmitted. Default output commonly lives under cover-image/topic/, including source material, prompts/, and cover.png, with alternative locations configurable.
Save the complete prompt before generating and verify the topic, title, and whitespace. Inspect both the full image and listing thumbnail: lettering, cropping, distinguishable colors, and whether the image accurately represents the article. Correct text through regeneration or a no-title version, keeping previous candidates for comparison.
When the direction is wrong, name the dimension to change: retain the concept and palette but reduce intensity, or simplify the text. Specific revision boundaries preserve accepted choices instead of restarting the entire concept on every generation.
Review a cover at both display sizes
Imagine an article titled “How teams hand off complex projects,” shown at the top of the article and in a three-column card grid. Ask the image to convey a handoff relationship rather than squeeze in every tool logo. Use one short line of title or no embedded text so the grid thumbnail stays legible. Set the main destination ratio with --aspect, control text with --text, and say whether a reference contributes only colors or also composition.
Inspect cover.png at full size and at the actual card width. Is the subject immediately clear, is anything important cropped, and does any title match the article exactly? An attractive image that implies a nonexistent software interface has a meaning problem, not just a style problem; revise prompts/cover.md. If only the small text is unreadable, try title-only or no-title instead of lengthening the image prompt. Keep candidates so you can choose the version that works in the real layout.
Define the elements and ratio for a wide cover
For a wide article-card cover, specify the required visual elements and display ratio as well as the title. In his own workflow, Jinchenma describes the elements and uses an approximately 2.35:1 ratio; without an image API, he asks the skill to save prompts and renders them in a separate tool. Review the concept before rendering, then place the finished image at the actual card width to inspect cropping and text. A saved prompt alone is not a publishable cover.