What’s new
- Generation and editing in one model: OpenAI positions it as its model "for fast, high-quality image generation and editing," with inpainting built in rather than needing a separate tool.
- Up to 16 reference images per edit call: whether you're generating from references or editing an existing image, a single call can take up to 16 input images.
- Resolution up to 4K, with transparent backgrounds: three resolution tiers (1K/2K/4K), and at 1K you can generate directly with a transparent background (PNG/WebP), skipping a separate cutout step.
- A wide spread of aspect ratios: 16 supported ratios, from square 1:1 to extreme banners like 21:9 and 9:21.
What it’s good for
- Product shots, icons and stickers that need a transparent background — generate straight onto a clean alpha channel instead of cutting one out afterward.
- Editing an existing image while preserving the rest — swap the copy on a product shot while keeping composition, lighting and packaging details untouched.
- Print or large-display work where resolution matters, using the 4K tier.
Prompting tips
- State the use case and composition up front: OpenAI's guide suggests naming the subject and intended use (a product photo, an ad, a diagram) first, then specifying composition, aspect ratio and where key elements should sit.
- Use whatever format is easiest for you to edit: there's no special syntax to memorize — plain prose, a labeled list, or tags all work, so pick whichever you can read and revise most easily.
- Ask for "photorealistic" explicitly when that's the goal: describe framing, materials and lighting too, since realism isn't the default assumption.
- Separate what changes from what stays the same, for edits: the recommended pattern is "change only X," then list everything to preserve — identity, geometry, layout, lighting, labels.
- Iterate one step at a time across multiple edits: feed the previous output back in as the next input, request a single change, and restate the details that must stay the same.
- Put on-image text in quotes: quote the exact wording you want rendered and describe its position and typography; spell out unusual words or brand names letter by letter if needed.
A product photo of a dark chocolate protein bar, 1:1, transparent background (PNG). Photorealistic,
soft overhead light, slightly glossy wrapper, visible nut pieces on the surface. The wrapper reads
"PURE FUEL" in bold sans-serif, centered near the top. Clean background, just the product and a soft
drop shadow, no extra decoration. Change only the wrapper text; keep composition, lighting and
materials the same.
Known limits
- Transparent backgrounds only work at 1K: setting background to transparent or opaque requires 1K resolution — 2K and 4K don't support it.
- Some aspect ratios can't go to higher resolutions: a 1:1 image can't be rendered at 4K; leaving aspect ratio on auto caps output at 1K; 5:4, 4:5, 3:1, 1:3 and 9:21 aren't available at 2K, and 3:1, 1:3, 9:21 aren't available at 4K either.
Using it in Nomi
On the generation canvas, add a image card, pick “GPT Image 2” in the model picker and choose a mode (Text to image, Image editing). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
