What tools let me edit my photos with GPT Image 2.6?

ChatGPT already had markup. The API has it now, so you can point at one area and leave the rest of the picture alone.

Markup is the edit. You mark the part that should change. The model is supposed to hold the character, product, and brand details outside that mark. Sunburst does that more steadily at medium or high quality. Flare is a fair first try when you mainly want to see the change quickly.

What the mark has to be

The mark is a transparent PNG the same size as the photo. It applies to the first image only. It cannot be combined with a mask. You can send up to four reference images with an edit. The first image is the one being changed. The others are references, such as a product or a brand detail you want kept.

A GIF or an SVG cannot enter that chain. Generate those formats on their own. If you need to change a still picture, start from the still.

How hard it should think

Quality sets the detail: quick, standard, or fine. Complexity sets how much planning the model does. For most edits, start at low complexity. Move to medium or high only when the picture has many parts or a strict brief.

On Image26, the box sends the prompt. The “Reference photo” control explains this rule. It does not upload a file. Do the marked edit in ChatGPT, or send the PNG through the API from your own server.