GPT Image 2 follows the brief to the letter
Other models interpret a brief; this one executes it. Write that three callouts go in the top left, middle right and bottom left and must not overlap, and that is what arrives. Which is why it is our model for graphics that explain something.
Rendered by this model
Specification
- Built by
- OpenAI
- Starts from
- a written brief · reference photos
- Size
- ≤ 3840 × 2160 px
- Reference photos
- 14
- Render time
- about 35s
- Cost
- 1 credit per image
What it is good for
- It respects placement
- A caption in the corner you named, a leader line reaching the right part of the product, nothing on top of anything else. Every other model turns that into a composition of its own.
- Polish captions without errors
- Short captions with diacritics come out correctly, in several places at once. For a graphic that describes features that is the whole idea.
- It refuses instead of handing back anything
- The only model on the roster that rejects a frame wider than 3:1 by itself. We would rather have a refusal than a differently shaped frame you would still pay a credit for.
When to pick something else
That refusal is also its limit: it stops at 3:1, so there is no ad strip coming out of it — that is a job for Nano Banana 2 or Seedream 5.0 Pro. It is the slowest image model here too, around thirty-five seconds a frame, and on a plain product photo with no callouts it adds nothing Seedream 4 will not do.
See every modelHow to brief this model
The things this engine handles differently from the rest. The habits that hold for every model are collected separately, in the help centre.
- State where each label goes. "Top left", "middle right" — this is the only model that genuinely holds to placement, so describe the arrangement rather than hoping for it.
- Write the labels exactly as they should appear. Short Polish labels — "stal nierdzewna", "600 ml" — come out here without error.
- Add that the leader lines must not cross. An explainer graphic with three pointers then comes out readable on the first version.
- Stay at 3:1 or narrower. A leaderboard strip is not possible here — Nano Banana 2 is for that.
- Expect the occasional refusal. This engine will sometimes return nothing rather than return something poor, and it questions a brief more often than the others — reword it in plainer terms, without brand names.
Where this model works
Questions about this model
Other models
- Nano Banana ProOne model for photos and for graphics. The most versatile on the roster.
- Seedream 5.0 ProExact pixel sizes and the widest frames. For formats the rest cannot hold.
- FLUX1.1 Pro UltraMaterial texture and believable light. For photos that should look like photos.
- FLUX.1 KreaA photo that does not look generated. Less gloss, more truth.
- Seedream 4Up to 4096 px on a clean background. The default for a listing's first photo.
- FLUX.1 DevFifteen seconds a frame. A solid middle between speed and quality.
- FLUX.1 SchnellSix seconds a frame. For checking an idea, not for finishing one.
- Krea 2 TurboEight seconds and a frame with a clear character. For mood shots.
- Qwen Image 3Correct Polish text and a clean layout. For graphics with a headline.
- Ideogram V3The best typography on the roster. For graphics where the text is the subject.
- Ideogram V4Ideogram's typography at up to 3840 px. For large creative.
- Nano Banana 28:1 and 1:8 frames the rest cannot do. For banners at extreme ratios.
- FLUX.2 ProDeep colour and glass with reflections. For creative from the top shelf.
- Z-Image TurboTen seconds to a clean composition. You add the text yourself, in the editor.
- Bria Product ShotYour photo, a new scene. The product stays exactly the same.
Try this model on your own product
Sign up without a card and get 40 credits. Switching model is one click, so you can send the same brief through several of them and keep the best frame.