Models

Generation and editing take an optional model parameter. Omit it and the request goes to google/gemini-2.5-flash-image; pass a modelId from the catalogue to route somewhere else.

curl -X POST "https://api.imagemcpserver.com/playground/generate" \
  -H "Authorization: Bearer $IMAGEMCP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "prompt": "Isometric illustration of a data centre, flat colours",
  "model": "google/gemini-2.5-flash-image"
}'

list_models returns the models currently marked live, highest priority first. It costs nothing, so an agent can consult it before deciding how to spend.

Not everything is model-selectable. These tools run on models chosen for the specific job, so the model parameter is ignored and the price is flat:

• remove_background and the cutout stage of generate_transparent_image — a dedicated background-removal model.
• upscale_image — a dedicated crisp upscaler.
• text_to_svg — a vector model, with an AI SVG fallback if the vector call fails.
• compress_image and convert_format — no model at all; these are local image operations, which is why they cost 1 credit and return in well under a second.

Two request fields are worth being precise about, because one reaches the model and one does not:

• aspectRatio is forwarded to the model as aspect_ratio. Common values are 1:1, 16:9, 9:16, 4:3, 3:4 and 21:9.
• style is recorded on the request and echoed back in result.style, but it never reaches the model — style direction belongs in the prompt.

• Iterate on prompts with a cheaper model, then switch once the prompt is settled — the composition usually carries over.
• Reach for text_to_svg rather than a raster model when the asset is an icon or a logo; you get editable markup instead of pixels.
• Read result.latency if speed matters to your product; it is the real wall-clock time of that call.
• Compare costPerReq before a large batch. On a hundred-image run the difference between models is the difference between a plan tier.