Editing a photo instead of generating a new one

Generating and editing answer two different questions. Generating answers "make me a picture of this". Editing answers "keep this picture, but change one thing about it". If your goal contains the words "this photo", you want the editing path — upload the image, and the request becomes an edit instruction rather than a fresh generation.

What changes once an image is uploaded

Everything else still applies. The style you pick is added as a look instruction ("render the edited result as a watercolour painting"), so turning a holiday photo into an illustration is a two-click job. The aspect ratio and the four switches work the same way they do without an upload, and the cost is unchanged: one credit, one image.

The prompt changes character, though. Without a photo you are describing a whole picture; with one you are describing a difference. "A cat on a sofa" is a generation. "Replace the sofa with a leather armchair" is an edit, and the shorter, more specific instruction wins.

The brush narrows the change further

Paint over the part you want changed, write what should happen there, and the instruction becomes "change only the marked area". This is how you replace a sky without touching the building underneath it, or remove one object from a table without regenerating the table. Without the brush, the whole frame is fair game; with it, the rest of the picture is protected.

You can mark several areas and give each one its own note, which is how a single pass handles "new sky here, remove the sign there". Marks are numbered and the numbering is what ties each note to its area, so keep the notes short and specific.

When to edit and when to start over

The edit styles page walks through the common jobs one by one, and Explore shows finished results with the prompt that made each one.