CommunityBildgithub.com

facepop-sticker-maker

Image skill. Generate complete Chinese sticker packs from an uploaded person, pet, or mascot reference image. Use when the user wants reference-image stickers, meme reactions, matching expressions and actions, transparent cutouts, individual PNG files, or packaged ZIP delivery. Do not use for ordinary single-image illustration requests.

Was ist facepop-sticker-maker?

facepop-sticker-maker is a Codex agent skill that image skill. Generate complete Chinese sticker packs from an uploaded person, pet, or mascot reference image. Use when the user wants reference-image stickers, meme reactions, matching expressions and actions, transparent cutouts, individual PNG files, or packaged ZIP delivery. Do not use for ordinary single-image illustration requests.

Funktioniert mit~Claude Code✓Codex CLI~Cursor
npx skills add https://github.com/yangxidong2018/facepop-sticker-maker/tree/HEAD

Installed? Explore more Bild skills: steipete/songsee, affaan-m/frontend-design, mattpocock/improve-codebase-architecture · View all 6 →

In Ihrer bevorzugten KI fragen

Öffnet einen neuen Chat, in dem dieser Agent-Skill bereits geladen ist.

Dokumentation

FacePop Sticker Maker

Turn one uploaded reference image into a coherent sticker pack and deliver individual transparent PNGs plus ZIP archives. Work entirely in the conversation; do not require or modify a website.

Required confirmation before generation

Do not start image generation until the user has seen and confirmed the sticker captions.

  1. Identify the usable reference image and subject.
  2. Present the proposed captions as a numbered list. Default to these 9 captions unless the user supplied another set: 欢迎、您辛苦了、比心、收到、爱你哦、感谢、拜托了、太棒了、赞
  3. Tell the user they may add, remove, reorder, or rewrite any caption.
  4. Also state the proposed style. Default to 搞笑手绘.
  5. Wait for explicit confirmation. A message such as “可以”“就这样”“生成” counts as confirmation.

If the user already supplied exact captions and explicitly asked to generate, treat that as confirmation; do not ask again.

Generation workflow

  1. Read references/style-and-identity.md.
  2. Read references/expression-mapping.md, then map every confirmed caption to a distinct pose, expression, and optional prop. Semantics must match even when the wording is custom.
  3. Generate a clean grid sheet without captions. Prefer a transparent background. Use one cell per confirmed caption in reading order; leave unused cells empty. Keep the same subject identity, clothing cues, rendering style, white sticker outline, scale, and line weight across the sheet.
  4. Inspect the sheet before slicing. Check identity, caption-to-action mapping, complete limbs/props, smooth eye corners, white outline, distinct poses, grid order, and empty unused cells.
  5. If the sheet passes, run scripts/package_sheet.mjs to create:
    • transparent 512×512 PNGs with captions;
    • matching transparent PNGs without captions;
    • one ZIP archive for each set;
    • manifest.json recording captions and file names.
  6. Inspect at least three individual stickers, including the most exaggerated smile or squint. Verify the ZIP archives.
  7. Return clickable links to both ZIP archives and the source sheet. Mention the number of stickers and any material limitation.

Typical packaging command:

node scripts/package_sheet.mjs \
  --source /absolute/path/source-sheet.png \
  --output /absolute/path/output-folder \
  --labels '["欢迎","您辛苦了","比心","收到","爱你哦","感谢","拜托了","太棒了","赞"]' \
  --cols 3 --rows 3 --name facepop

The packaging script needs Sharp. It first tries the installed Node package, then the Codex desktop bundled runtime. If neither is available, report the missing dependency rather than silently changing the deliverable.

Retry rule

Allow at most 2 regeneration attempts after the initial generation.

Before each retry, tell the user briefly:

  • what failed or did not meet the confirmed requirement;
  • why regeneration is necessary;
  • which constraint will be strengthened.

Use a retry only for a material defect such as incorrect identity, wrong action mapping, clipped subject, missing white outline, unwanted wrinkles, duplicated cells, broken text order, or unusable transparency. Do not retry merely for harmless stylistic variation. After 2 unsuccessful retries, stop, deliver any usable partial result, and explain the remaining defect.

Output rules

  • Default count: 9. Respect a different confirmed count.
  • Product wording: use 搞笑, never rename the style to 沙雕 unless quoting legacy material.
  • Default visual style: hand-drawn funny illustration, not photorealistic.
  • Captions are added during packaging, not baked into the generated source sheet. This ensures identical with-text and no-text editions and accurate Chinese text.
  • Preserve the generated source and do not delete user files.
  • Use a fresh output directory for each run so packs do not overwrite one another.
  • Do not upload, publish, or send the results outside the conversation without separate authorization.

Verwandte Skills