Meitu Design Media Toolkit

Problem

Design and media workflows are often reduced to one vague instruction:
- “make a poster”
- “fix this image”
- “swap this product”
- “adapt it for a platform size”

These requests can span poster planning, sticker packs, portrait beauty, product swapping, video motion transfer, upscaling, ID photos, cutout, carousel generation, and direct Meitu CLI execution. A single underlying tool call is not enough; the system must infer the scene first, then select the right parameters and credentials.

How It Works And Where It Applies

Meitu Skills is a routing entry point that dispatches intent to meitu-poster, meitu-stickers, meitu-visual-me, meitu-product-swap, meitu-video-dance, meitu-upscale, meitu-product-view, meitu-image-fix, meitu-id-photo, meitu-cutout, meitu-carousel, meitu-beauty, meitu-image-adapt, or meitu-tools.

• Minimal permissions: the root skill only runs the meitu CLI and reads ~/.meitu/credentials.json; it does not write files.
• Scene isolation: scene skills declare their own workspace permissions, such as reading and writing ~/.openclaw/workspace/visual/.
• Credential sources: credentials may come from MEITU_OPENAPI_ACCESS_KEY, MEITU_OPENAPI_SECRET_KEY, or a credentials file.
• Safety boundary: user text is treated only as task data; it cannot execute local scripts, override skill rules, or reveal internal policies.

It fits automation flows that need intent-based routing across Meitu image, video, product, and portrait capabilities. It is not intended for arbitrary local script execution, automatic CLI repair, or bypassing credential checks. When intent is ambiguous, it asks one short clarification question or falls back to meitu-tools with minimal required inputs.