OpenAI · Images · premium
GPT Image 2
OpenAI's image model. The best instruction-follower here, with a quality dropdown that swings the price elevenfold.
What it costs
| Quality · auto | 11 credits |
| Quality · low | 1 credit |
| Quality · medium | 4 credits |
| Quality · high | 11 credits |
What it takes
| Aspect ratios | 1:1 3:2 2:3 4:3 3:4 16:9 9:16 |
| Reference images | 7 |
When to reach for it
This is the model to reach for when the prompt is a set of instructions rather than a description: three named objects in stated positions, a sign that has to read as particular words, a layout with a specific hierarchy. It holds compositional intent through a long prompt more reliably than anything else in the catalog, and that is what the premium tier buys.
The quality setting deserves attention before anything else. On “auto”, which is the default, a generation is 11 credits. On “low” it is 1. That is the same model reasoning about the same prompt at a tenth of the price, and for the phase where you are still finding the composition, “low” tells you everything you need to know. Switch to “high” when the layout is settled and only the render quality is left.
Seven references, seven aspect ratios. Ratio coverage is narrower than Nano Banana's, so if the shape is unusual this is not the model, however well it follows instructions.
Made with it
Call out the parts that matter
Your words, printed exactly as you wrote them.
Make it a moving page
Nine collage scenes, one shared look.
Lay it out as one page
A whole recipe poster, typeset in one image.
Explain it in fifteen seconds
Four flat-vector scenes, cut to the voice.
In the arena
Ranked #1 with a rating of 1591. 2 of 4
Week 36, 2026 — Arena week 36: every model can spell nowCompared with
- Nano Banana 2 vs GPT Image 2
GPT Image 2 when the prompt is a set of instructions; Nano Banana 2 when the shape is unusual or the volume is high.
Guides that cover it
- Grok Imagine Video 1.5: the complete prompting guide
One picture in, up to fifteen seconds out, with sound you cannot turn off. Nine rules, three pipelines and the failures worth knowing about first.
- P-Video: the complete guide to Pruna's four models
Four models, one family: a cheap generator and three performance rigs. What each one takes, what a take really costs, and the failures worth knowing first.
- Seedance 2.5: the complete prompting guide
Thirty seconds of video with sound from one written brief. Seven rules, three pipelines and the failures worth knowing about before you spend anything.
- Veo 3.1: the complete prompting guide
Eight seconds of video that writes its own soundtrack. Three audio channels, eight rules, four pipelines and the failures worth knowing about first.
- Which image model to reach for, and when
Twelve image models, from half a credit to twenty-one. What each one is for, where the resolution setting changes the price, and when the cheap tier is right.
GPT Image 2 is one pick in the composer.
Every model in this catalog runs from the same prompt bar, priced in the same credits. New accounts start with some to spend.
Start picturing