Campaign layouts
explore different crops and compositions for a single visual direction.
Image generation & editing
Generate and revise images across a wide range of formats.
Loading model settings…
Your prompt, settings, and references stay together in a new Canvas.
01 / Meet the model
GPT Image 2 combines prompt-led creation with reference-guided editing. Use it when one visual needs several compositions: a square product tile, a portrait campaign image, and a wide banner. Describe what should change and what should stay before you create the next version.
02 / Choose it for
explore different crops and compositions for a single visual direction.
bring product, subject, or style images into the brief rather than describing everything from scratch.
choose an output size that lets you inspect small elements before deciding on a final asset.
03 / Before you create
Model settings. Availability, input mode, and references can narrow the choices offered in Canvas.
04 / A closer look
Establish one composition first, then generate each target ratio separately. A new ratio can change framing; inspect text, hands, and product details in every result rather than assuming an automatic crop preserves them.
Text prompt · optional reference images. The input controls above show the options available for your selected mode.
Auto · 1:1 · 3:2 · 2:3 · 4:3 · 3:4 · 5:4 · 4:5 · 16:9 · 9:16 · 2:1 · 1:2 · 3:1 · 1:3 · 21:9 · 9:21. 1K · 2K · 4K.
Start with the format you intend to publish, then choose the available output settings. Changing inputs can change the choices. Review the current Gem estimate before Generate; availability and cost can vary.