Your product in a new setting

When to use it

You have a photo of the product, from a designer, a phone or an earlier render, and you need it somewhere else: on a café table for the landing page, on a desk for an ad, in a kitchen for a blog header. Pillow can cut it out and paste it on a painted table, but the light won't match and it looks pasted. In our side-by-side test the café edit was the best single result across three runs, and a blind judge preferred it over the code version (how we tested).

What to ask your agent

Put our coffee bag photo (assets/bag.png) on a sunny café table by a window. Keep the bag, the cup and the beans exactly as they are.

What your agent sends

The hosted server can't read your disk, so a local file goes up first. The agent calls vvgen_create_upload with the file name, runs the single-use upload command it gets back, and receives an asset id. In the Claude app, attach the photo: Claude uploads it itself once you allow api.vincevangen.com (how), otherwise it gives you the link to open, pick the photo, and gets the id when the upload finishes. A public https:// URL or the id of an earlier result works directly.

Then vvgen_edit_image. The call from our test, word for word:

{
  "images": ["a_eb7d0b1fef"],
  "prompt": "Place the coffee bag, the teal cup and the beans on a sunny wooden café table by a window, with warm morning sunlight and a softly blurred café interior behind; keep the bag, label band, cup and beans exactly the same in shape, colour and position, keep everything else exactly the same.",
  "seed": 11
}

Lead with the change, then say what must stay: "keep everything else exactly the same". The first image is the canvas and keeps its size. Up to four images can go in one edit; the others are sources the prompt names as "image 2" and "image 3", for example a logo to put on the bag.

What comes back

The edited file, usually a JPEG (add "post": ["png"] for a PNG), a preview, the credits and the model's licence. Its record lists the photo it was made from by id and SHA-256. This edit was made by the open-source engine's local default edit model at the time (FLUX.2 [dev]); the hosted Edit tier runs FLUX.2 [pro].

Ask your agent to compare the preview with the original: label shape, colours, the number of beans. If something drifted, a second edit that names only that thing works better than a longer first prompt.

Each extra image in the same edit adds to the price; the quote your agent gets before the call shows the total.

In Cursor and Codex

Same tools: see the Cursor and Codex quickstarts. In Codex, attaching the photo to the chat is not enough for an MCP tool: the agent still uploads it with vvgen_create_upload.

claude mcp add -s user --transport http \
  vincevangen https://api.vincevangen.com/mcp