Edit image

Take an image the user already has and change it: add, remove, or replace objects, recolor or relight, blend elements, restore, inpaint a masked region, or outpaint to extend the canvas. The work starts from their pixels and keeps everything they did not ask to change. To make a fresh image from a prompt instead, that is a generation task, not this one.

Inputs to collect

• The source image. Required. URL, base64, data URI, or a UUID from a previous generation or the Image Upload API. (Ask only if none provided.)
• The instruction: what to add, remove, replace, recolor, relight, blend, or extend. One plain sentence is enough.
• A region, if the edit is local. For removal or inpaint, a binary mask (white = the area to act on, black = keep). For outpaint, the per-side pixel extension.
• What must stay untouched (the subject, the framing, the palette), so the edit stays surgical.

Models

Route by the kind of edit. Confirm the live model and inspect its schema via runware-models + runware-run before calling. Never hardcode a stale choice.