I am looking for an mcp that will look at images and separate their layers then make new images
2 people need this · 2 named a price · around $0.00335 per call
Builders
0 of 3 slots takenUp to 3 builders can work on this at once. Every build that passes verification gets published, so they compete on price and quality rather than on who finishes first.
The room
Anyone can readWhere the person who asked and the builders working on it talk. Open to read so you can see whether this is moving before you back it.
Nothing here yet. It starts when a builder takes a slot.
Only the requester and the builders who have taken a slot can post here.
The spec
- name
- Separate image into layers and export each
- tool
- separate_image_layers
- input
- image_url: string, or image_base64: string; layer_types: array of strings (e.g. ["foreground", "background", "shadow", "text"]), optional; output_format: string ("png" or "webp"), optional, default "png"
- output
- layers: array of objects, each containing layer_name: string, layer_base64: string (PNG with alpha channel), bounding_box: object {x, y, width, height}; layer_count: integer
- delivery
- Either
Examples
0 of 3 confirmedWhat a build has to handle. For now the builder and the requester agree these between them. Automatic testing against a held-back set is coming, and this page will say when a request is covered by it.
You give it
A product photo of a white trainers on a plain grey studio background. No layer_types specified.
It should give back
At least two layers returned: one containing the shoe with a clean alpha cutout and no grey fringing, and one containing the background. Each is a valid PNG with an alpha channel.
Drafted, not yet confirmed by the requester.
You give it
A street photograph containing a person in the foreground, a building mid-ground, and a sky background. layer_types: ["foreground", "midground", "background"].
It should give back
Exactly three layers returned matching the requested names. The person layer isolates the figure with no building pixels bleeding in. The sky layer contains only sky. All layers are the same pixel dimensions as the input.
Drafted, not yet confirmed by the requester.
You give it
A flat graphic with overlapping coloured text on a patterned background, submitted as a base64-encoded PNG. No layer_types specified.
It should give back
Text elements and background pattern are returned as separate layers. Text glyphs retain sharp edges with no anti-aliasing artefacts from the pattern bleeding through. Layer count is at least 2.
Drafted, not yet confirmed by the requester.
How it will be checked
By you, the requester
This build could not be run automatically, so you decide whether it does the job.