SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsmeitu-stickers

meitu-stickers

General & other

View repository

Updated 10-16 06:46

About this skill

Meitu Stickers

Overview

Generate multi-style 2x2 grid meme stickers (2x2 grid) from user-uploaded photos, split into 4 independent stickers, and optionally convert specified stickers or all of them into animated GIF meme packs. Supports four built-in styles: Q-version, 3D clay, pixel style, and Emoji, as well as user-defined styles.

Dependencies

  • tools: meitu-cli — Meitu AI Open Platform CLI (image generation/editing, video generation, format conversion)
    • Install: npm install -g meitu-cli@latest(package name meitu-cli)
  • credentials: Meitu AI Open Platform API credentials
    • Environment variables:MEITU_OPENAPI_ACCESS_KEY / MEITU_OPENAPI_SECRET_KEY
    • Or configuration file:~/.meitu/credentials.json
    • Preferred environment variables:MEITU_OPENAPI_ACCESS_KEY / MEITU_OPENAPI_SECRET_KEY
    • Or preset credentials file:~/.meitu/credentials.json
    • If manual initialization of local credentials is needed, you may explicitly run meitu config set-ak --value "..." + meitu config set-sk --value "..."(will write to local file)
    • Verify:meitu auth verify --json
  • workspace (optional): {OPENCLAW_HOME}/workspace/visual/
    • Not found → skip all knowledge reads, skill works without it

Path alias: Below, $VISUAL = {OPENCLAW_HOME}/workspace/visual/

Core Workflow

Preflight → [Context] → Execute → Refine → Deliver → [Record]
              ↑ 创意型任务执行                   ↑ 项目模式时执行
              ↑ 工具型任务跳过                   ↑ 一次性模式跳过

Preflight

  1. meitu --version → if not installed, prompt npm install -g meitu-cli@latest
  2. meitu auth verify --json → if credentials are invalid, guide configuration
  3. Detect mode: cwd has openclaw.yaml → project mode; else → one-off Check $VISUAL directory → determine capabilities can_record = cwd has openclaw.yaml AND $VISUAL exists (if either is missing, false)
  4. output_dir resolution (MUST be completed within Preflight): Resolve output_dir: openclaw.yaml → ./output/ | else → $VISUAL/output/meitu-stickers/ mkdir -p {output_dir}

Context (executed for creative tasks + project mode)

mode = one-off → skip this step, go directly to Execute.

User explicitly references brand assets AND $VISUAL exists → read only as needed $VISUAL/assets/

mode = project → execute step by step:

  1. Read ./DESIGN.md
  2. Extract Context References → try to read $VISUAL/assets/ or $VISUAL/rules/ global assets → if unreadable, use inline fallback values in DESIGN.md
  3. Read quality.yaml → global.md → scenes/{type}.md (type is read from openclaw.yaml's project.types(array takes priority) or project.type , both skip if missing) → quality forbidden list filters generated elements, preferences enhance creative direction; design is null → create after Execute

Execute

Classify Input

Scenario Condition Action
A — Single clear subject photo User uploads 1 image containing an identifiable subject (person/animal/character/building/product/food) → Select Style
B — Multiple photos User uploads > 1 image → Composite → Select Style
C — Already in meme pack style User's image is already in some meme pack style → Select Style, treat as refinement/editing request
D — No image uploaded No image Reply with prompt (see below), wait for upload

Scenario D reply: "Please upload a photo, and I'll help you generate meme stickers~" If it contains multi-image keywords such as "a group"/"a set" → additionally prompt "If you need a group photo effect, please select and upload multiple images at once"

Composite Multiple Photos (Scenario B only)

meitu text-to-image \
  --skill_name skill_meitu-stickers \
  --image_list "<image_url_1>" "<image_url_2>" \
  --prompt "将多张图片主体物合成一张合照（如第一张照片的人物与第二张照片人物合照），{user_pose_requirement}，保持主体物样貌相似度不变，保持图片风格不变" \
  --json --download-dir {output_dir}/composite
  • {user_pose_requirement}: if the user specified a pose/interaction, add it; otherwise omit this clause.
  • From downloaded_files[0].saved_path Read the composite result and use it as the source image for Generate Sticker Grid.

Select Style

Style recognition decision table:

User keyword Style ID Description
Chibi, chibi, big head sticker, cute chibi chibi Big head small body, rounded features, exaggerated expressions
3D, clay, clay, clay sculpture, clay clay 3D clay/plasticine texture, soft and rounded, matte texture
Pixel, pixel, 8bit, retro game pixel Pixel art style, crisp pixel edges, limited palette
emoji, emoji, round face emoji Thick outlines, flat colors, simplified features, round face
Other specified styles (e.g. "watercolor", "cyberpunk") custom Build prompt based on user description
No style specified — Proactively ask (see below)

When no style is specified, ask: "Which style of sticker do you want? I support these built-in styles: 🎨 Chibi — Big head cute style 🧸 3D Clay — Three-dimensional clay texture 👾 Pixel — Retro game art style 😊 Emoji — Round face flat style

You can also tell me any style you want, such as "watercolor style" or "cyberpunk"~"

Generate Sticker Grid

Prompt template: See references/prompts.md。

  • Select the corresponding style section (## chibi / ## clay / ## pixel / ## emoji / ## custom）
  • Concatenate ## GRID_CONSTRAINTS Ensure the generated result can be image-grid-split correctly split

Style → command decision table:

Style ID Command Reason
chibi text-to-image --image_list Reference a real person image for artistic stylization (big head chibi)
clay text-to-image --image_list Reference a real person image to generate 3D clay style
pixel text-to-image --image_list Reference a real person image to generate pixel style
emoji text-to-image --image_list Reference a real person image to generate flat emoji style
custom(Art style type: watercolor/cyberpunk/oil painting, etc.) text-to-image --image_list Under 2.0.6, stylization prioritizes reference-image-based generation
custom(Realistic type: portrait/ID photo, etc.) text-to-image --image_list Preserve character features, but avoid relying on the old image-edit style model
custom(Uncertain) text-to-image --image_list Uniformly use reference-image generation to reduce routing divergence

Note: The current CLI already supports image-edit --model nougat as an explicit high-quality editing path; however, sticker scenarios still uniformly prioritize text-to-image --image_list, because four-grid stickers are better suited to the reference-image generation pipeline, with more stable style and composition control.

Generation command:

meitu text-to-image \
  --skill_name skill_meitu-stickers \
  --image_list "<source_image_url>" \
  --ratio 1:1 \
  --size 2K \
  --prompt "{STYLE_PROMPT} {GRID_CONSTRAINTS}" \
  --json --download-dir {output_dir}

{STYLE_PROMPT} Responsible for specifying the style type (chibi / clay / pixel / emoji / custom).

Error fallback strategy (L1-L5):

Level Operation Specific content
L1 Remove low-priority modifiers Remove descriptive modifiers from the style prompt (e.g. "matte clay texture with subtle highlights"), keep core style words + GRID_CONSTRAINTS
L2 Simplify style description Reduce the style prompt to one sentence: e.g. chibi → "2x2 grid of chibi stickers from this photo, 4 expressions" + GRID_CONSTRAINTS
L3 Remove optional inputs N/A (source image is a required input)
L4 Minimize to core elements prompt → "2x2 sticker grid, 4 stickers, {style} style, white background, well separated"
L5 Stop and report error 2 consecutive failures → report error to the user, with code and hint

Show the 2x2 grid to the user, and wait for confirmation before splitting the image. Do not directly execute grid-split. → Enter Refine

Refine

Phase 1: Confirm the 2x2 grid

Show the generated 2x2 grid, explain the style used, and ask: "This is the generated {style_name}-style 2x2 grid sticker pack. Are you satisfied with the overall effect? If so, I'll help you split it into 4 separate stickers~"

Wait for the user's reply:

Feedback classification and handling:

Feedback type Example Handling method
Change style "Switch to pixel style", "Try 3D" Go back to Select Style → Generate Sticker Grid
Adjust expressions "The second expression is too exaggerated", "Make it cuter" Adjust the expression description in the prompt → Generate Sticker Grid
Adjust style details "Make the colors more vivid", "Make the lines thicker" Append prompt modifiers → Generate Sticker Grid
Image splitting issues "The stickers are too close together", "There is overlap" Strengthen GRID_CONSTRAINTS spacing → Generate Sticker Grid
Satisfied "OK", "Sure", "Satisfied" → Enter Phase 2

It is recommended to iterate at most 3 rounds; after that, proactively suggest adjusting the direction or splitting the requirements.

Phase 2: Split the image + confirm stickers

After the user confirms the 2x2 grid, execute the image split:

meitu image-grid-split --image_url "<grid_image_url>" --json --download-dir {output_dir}/split --skill_name skill_meitu-stickers

If image-grid-split returns fewer than 4 images（grid spacing insufficient for detection）:

  1. Re-generate 2x2 grid with stronger spacing prompt — append: "Ensure each sticker is clearly separated with at least 25% white space between them. Each sticker must be a completely independent illustration with no visual connection to adjacent stickers."
  2. Retry image-grid-split. If still < 4 → deliver grid image as-is and inform user.

Show all 4 separate stickers at once(after downloading all images, show them all in the same reply, do not split into batches), and ask: "All split~ Would you like to convert one of them or all of them into animated stickers (GIF)? Just tell me the numbers (e.g., 1, 3) or say "all". If you don't need it, just say you're satisfied and I'll help you save them~"

Wait for the user's reply:

  • No GIF → Enter Deliver
  • Specify GIF conversion(one / multiple / all) → Execute GIF conversion (below) → Show results → Enter Deliver

GIF conversion process (execute sequentially, one by one):

a. Image to Video:

meitu image-to-video \
  --skill_name skill_meitu-stickers \
  --image_list "<selected_sticker_url>" \
  --prompt "{style_name} character performing a simple animated expression, loopable motion" \
  --json \
  --download-dir {output_dir}/video

Wait for async result (CLI automatically polls, video downloaded to {output_dir}/video）。

b. Show the video to the user(already downloaded to {output_dir}/video), and let the user preview the animation effect.

c. Video to GIF (use the video file downloaded locally by image-to-video):

meitu video-to-gif --video_url "{output_dir}/video/<video_task_id_file>" --prompt "transparent animated sticker" --json --download-dir {output_dir}/gif   --skill_name skill_meitu-stickers

If user selects "all", process all stickers sequentially (one at a time: image→video→show video→gif, then next). If user selects multiple (e.g., "1 and 3"), process the specified ones sequentially.

Deliver

The file is already in output_dir(used during generation --download-dir), only rename is needed:

mv {output_dir}/split/{task_id_file} {output_dir}/{YYYY-MM-DD}_{style}-sticker-{n}.{ext} Animated GIF → mv {output_dir}/gif/{task_id_file} {output_dir}/{YYYY-MM-DD}_{style}-sticker-{n}-animated.{ext}

Record (executed in project mode)

can_record = cwd has openclaw.yaml AND $VISUAL exists (if either is missing, false) → skip when false. In one-off mode, feedback is only valid for the current conversation. Exception: in one-off mode, if the user repeatedly expresses the same preference → Agent MAY propose writing to $VISUAL/rules/quality.yaml

User approved style → read $VISUAL/memory/observations/observations.yaml → scan similar key → merge or append → write back. len(projects) >= 2 → propose promotion (non-blocking). Non-blocking promotion proposal (mention at the end of the reply, without interrupting the main flow) → confirmed → write target + delete observation entry

User rejected ("不要 XX") → has openclaw.yaml → ask scope → project: ./DESIGN.md Constraints / universal: quality.yaml no openclaw.yaml → current task only

No feedback → skip entirely.

Output

  • 4 independent meme stickers (from grid split), style consistent with user's choice
  • Optional: animated GIF meme (for selected sticker(s))
  • File naming:{YYYY-MM-DD}_{style}-sticker-{n}.{ext}（e.g., 2026-03-22_chibi-sticker-1.jpg、2026-03-22_pixel-sticker-3-animated.gif）