Feature

API

Magic Prompt for image edits

With input images attached, magic_prompt now rewrites your request as an edit instruction and leaves the rest of the image alone.

Magic Prompt used to rewrite every prompt as a full scene description, even for edits. The model then redrew the whole picture and made up details it had never seen. Now a request with inputs gets a rewrite written for editing. It says what to change and asks the model to keep everything else.

const task = await mynth.image.generate({
  model: "black-forest-labs/flux.2-pro",
  prompt: "pls make the hair red",
  inputs: ["https://cdn.example.com/portrait.png"],
  magic_prompt: true,
});

console.log(task.magicPrompt?.positive);
// "Change the hair to red. Keep everything else unchanged."

This covers edits and generating from reference images. The rewrite knows each image's role, source or reference, so it can tell "edit this picture" apart from "use this style". Mynth never sends the image itself, its URL, or its filename to the rewrite. Requests without inputs work as before. See enhance prompts and image to image.