image-generate already returns a hosted PNG from a prompt. The listing now says that in the terms an agent is likely to search: AI image generator, text-to-image API, and prompt-to-PNG API.

The call itself hasn't changed. Send prompt with optional dimensions, tier, model, and seed. The response still includes the image URL, resolved model, final dimensions, prompt, seed, and timing.

Keep those response fields with an approved asset. You can rerun the same request later or change one control to make a deliberate variant. The same copy is available in the Synthforge MCP package, so an agent can find the repeatable image-production route from either catalog.