Q版动漫扁平几何插画

Chibi Anime Flat Geometric Illustration

Create one original chibi anime character illustration with crisp geometric shapes, a soft-vinyl finish, and a repeatable visual identity.

Style: Bold Grid Chibi. Playful and composed, with a clear character silhouette grounded by one saturated canvas color. Follow references/style-guide.md exactly.

Use this skill when

• Creators developing a recognizable avatar or mascot
• Illustrators exploring a consistent chibi character direction
• Indie teams making original character-led identity art

Do not use

• Photorealistic portraits, painterly scenes, or detailed environmental concept art
• Exact replicas of existing copyrighted characters, real people, or branded mascots
• To reproduce another company's brand, logo, product UI or copyrighted characters.

Inputs

• Character role, personality, and intended audience
• Distinctive appearance, clothing, and one signature detail
• Preferred canvas ratio and intended placement
• Optional palette or existing series reference description

If something essential is missing, ask one short question before starting. Never invent facts, numbers, quotes or customer names; mark gaps as [needs source].

Workflow

1. Collect the brief fields listed in references/prompt-framework.md. If any text will appear in the image, confirm it letter by letter and freeze it.
2. Build the prompt with references/prompt-framework.md, following references/visual-system.md for subject, composition, light, palette and the text policy. Write it to prompt.txt.
3. Check that the generation key is configured: SKILLLOOM_API_KEY and SKILLLOOM_BASE_URL (run npx skillloom doctor --write-env if not).
4. Tell the user the image is paid from their prepaid key (see the cost note) and get explicit approval.
5. Generate one image. The task id is saved first, so re-running the same command resumes instead of paying twice:

``bash node scripts/generate.mjs --prompt-file prompt.txt --output image.png --task-file image-task.json --confirm-paid ``

1. Run the validator and fix every problem it reports:

``bash node scripts/validate-image.mjs image.png ``

What makes this excellent

• Translate the brief into a memorable silhouette and one signature detail before adding clothing or props.
• For a series, preserve head-to-body ratio, eye spacing, light direction, and ground color; vary only brief-specific traits.
• Choose a pose that reads at small avatar scale, with hands and accessories separated from the body silhouette.
• Use crisp flat color boundaries; reserve soft shading for a few broad vinyl-like highlights and the grounding shadow.
• Keep the character centered with enough clear margin for circular profile crops unless the user specifies another placement.
• Treat any supplied character reference as descriptive guidance; do not reproduce protected characters, brands, or identifiable people.

Output contract

• One PNG at 1:1, at least 1000px wide, following the visual system; any approved text spelled exactly.
• No logos, watermarks or real people's likenesses unless the user owns them.
• A hand-off note with the final prompt, the task file path and the validator output.
• Generate one original illustration at the requested ratio and minimum width.
• Keep the character silhouette and signature details legible at thumbnail scale.
• Deliver a clean image without lettering, watermark, logo, or mockup framing.

Reference demo and agent run

The preview image was generated from the sample brief in scripts/image-spec.json. In a real run the agent fills the brief from the user's request, generates with the user's key, checks the result against the output contract (required text, ratio, no extra words) and regenerates when it misses.

Failure rules

• If the API reports a low balance or a limit, stop and tell the user; never switch providers or keys silently.
• If text is misspelled or the composition misses the visual system, ask before generating again. Never retry automatically; every attempt is paid.
• Never print or store the API key in files you hand over.