The GPT Image 2 page on Fooocus.one is built around a simple loop: describe the image you want, add art direction such as composition, lighting, palette and camera framing, then generate and refine with follow-up instructions. It supports aspect ratios from square and vertical to ultra-wide cinema formats, and output resolutions up to 4K, with a quantity selector for producing multiple variations in one run.
The emphasis is on editing rather than one-shot generation. Instead of rewriting an entire prompt after a disappointing result, you keep the concept and adjust specific elements — swap a studio backdrop for a warm beige one, make a product label more legible, or shift an illustration toward photorealism. The page also documents prompt tips for avoiding cluttered props, extra hands and distorted text, which are persistent failure modes in image models.
Fooocus.one bundles this mode alongside other image and video models such as Nano Banana, FLUX2, Z-Image, Seedream and Veo-style video generators, plus helper apps for photo restoration, background removal, face swap and upscaling. Because it is a third-party aggregator, model availability and naming are set by the site rather than by any official model vendor.