Edit an image with a prompt instead of a mask

Send a source image (a URL or inline data, as JSON or multipart) and a prompt describing the change; the edited image comes back the same way a generated one does. The card lists only the models the editing route accepts — several image models generate but do not edit.

The OpenAI-family models also take a mask for inpainting a single region; the Google-family models edit from the prompt alone. Both accept more than one source image for a fusion edit, each up to its own limit, and a mask cannot be combined with multiple images.

In Claude Code, Claude Desktop, Cursor, Windsurf, Cline and any other MCP host, this is the blockrun_image tool of the BlockRun MCP server — the agent calls it like any other tool and the result comes back in the conversation. In OpenClaw, Codex and Hermes, ClawRouter exposes the same catalog. In your own code, the Python and TypeScript SDKs wrap the endpoint above with types; any HTTP client works too. Every harness, with its setup →

No account and no API key stand between an agent and the first call: the request is quoted in dollars and the agent's client pays it. A team that prefers keys gets one, with a monthly invoice.

Image editing APIs come in two shapes. Consumer-grade services — Photoroom, Clipdrop, remove.bg — expose fixed operations: cut out the background, upscale, relight. Model APIs — OpenAI's edits endpoint, Google's image models — take an instruction in plain language and change whatever it describes. This page is the second kind, across more than one model, so a prompt like "replace the sky and keep the building" works without choosing an operation first.

The models on the card differ in what they accept. OpenAI's take an optional mask for inpainting one region; Google's edit from the prompt alone and are the better choice when there is no mask to draw. Both take more than one source image for a fusion edit. The card lists only the models the editing route accepts — several image models generate but do not edit, and they are not here.

Billing is per edit at the same per-image rate as generation on the model you pick, quoted before the request runs, and the edited image is returned on a URL that does not expire.