How do I use GPT Image 2.6 to generate and edit images?

Generate the picture first. Then mark the one spot that is wrong, and leave the rest alone.

On this page the order is the same every time. Choose the file, choose the model, write the picture, then generate. An edit is a second step, after you have a still image worth keeping.

Generate

1. Choose Image, GIF, or SVG.
2. Choose Flare when you want the picture fast. Choose Sunburst when the next version should still look like this one.
3. Write the subject, the setting, and what must not change. A few words is enough to start.
4. Set the shape. Square is 1024×1024, wide is 1536×1024, portrait is 1024×1536, and poster is 1024×1280.
5. Set the detail: Quick, Standard, or Fine. That is low, medium, or high quality.
6. Open More if the scene is complicated. Planning runs from Simple to Thorough. Simple is the right first edit.

A GIF from this box is 9 frames at 200 milliseconds. An SVG is one file. The page does not stream a half-finished image while it works.

Edit

Keep the still you want, then mark only the part that is wrong. In ChatGPT that mark already exists. Through the API, send a transparent PNG the same size as the photo, on the first image only, and do not combine it with a mask. Up to four references can come with that request. GIF and SVG stay out of this step.

If the second picture has to match the first, switch to Sunburst at standard or fine detail before you edit again.