Ratios and sizes
The image sizes you can pass in the size field of the Cheapest Imagen API: up to 19 aspect ratios at 1K, 2K and 4K, and custom sizes.
Pass one of these values in size. Every resolution costs the same: the model’s price per image.
The plain gpt-image-2, gpt-image-2.5 and gpt-image-2.6 are 1K only and use the first column; every other model uses the last three.
| Ratio | GPT Image 2 / 2.5 / 2.6 (1K) | Other models 1K | 2K | 4K |
|---|---|---|---|---|
| 1:1 | 1024x1024 | 1024x1024 | 2048x2048 | 2880x2880 |
| 3:2 | 1536x1024 | 1536x1024 | 2048x1360 | 3504x2336 |
| 2:3 | 1024x1536 | 1024x1536 | 1360x2048 | 2336x3504 |
| 4:3 | 1443x1090 | 1152x864 | 2304x1728 | 3264x2448 |
| 3:4 | 1090x1443 | 864x1152 | 1728x2304 | 2448x3264 |
| 16:9 | 1672x941 | 1280x720 | 2048x1152 | 3840x2160 |
| 9:16 | 941x1672 | 720x1280 | 1152x2048 | 2160x3840 |
| 5:4 | 1408x1120 | 1120x896 | 2240x1792 | 3200x2560 |
| 4:5 | 1120x1408 | 896x1120 | 1792x2240 | 2560x3200 |
| 21:9 | 1920x832 | 1456x624 | 2912x1248 | 3840x1648 |
| 9:21 | 832x1920 | 624x1456 | 1248x2912 | 1648x3840 |
| 2:1 | 1792x896 | 1536x768 | 3072x1536 | 3840x1920 |
| 1:2 | 896x1792 | 768x1536 | 1536x3072 | 1920x3840 |
| 3:1 | – | – | 2048x688 | 3840x1280 |
| 1:3 | – | – | 688x2048 | 1280x3840 |
| 4:1 | – | 2048x512 | 4096x1024 | 8192x2048 |
| 1:4 | – | 512x2048 | 1024x4096 | 2048x8192 |
| 8:1 | – | 3072x384 | 6144x768 | 12288x1536 |
| 1:8 | – | 384x3072 | 768x6144 | 1536x12288 |
1K, 2K and 4K work on nano-banana-2, nano-banana-2.1, nano-banana-pro, gpt-image-2-adobe, gpt-image-2.5-flare, gpt-image-2.5-sunburst, gpt-image-2.6-flare and gpt-image-2.6-sunburst. The Nano Banana stable channels make the resolutions in their name: nano-banana-2-1k and nano-banana-pro-1k 1K, nano-banana-2-2k 2K, nano-banana-2-4k and nano-banana-pro-4k 4K, nano-banana-pro-2k 1K or 2K. With size auto a single-resolution model uses its own resolution.
Which ratios a model offers:
- The first ten (1:1 to 21:9) work on every model.
- 9:21, 2:1 and 1:2 work on the GPT models; 3:1 and 1:3 on the GPT models at 2K and 4K.
- 4:1, 1:4, 8:1 and 1:8 work on
nano-banana-2andnano-banana-2.1only.
Custom sizes
gpt-image-2-adobe and gpt-image-2.5-flare also take a width and height of your own, such as 1600x896. Both edges must be multiples of 16, the longer edge at most 3840 px and at most three times the shorter, and the image 655,360 to 8,294,400 pixels. A custom size is priced like any other.
Exact pixels
The plain gpt-image-2, gpt-image-2.5 and gpt-image-2.6 may return slightly different pixels from the size you asked for, at the same ratio: in a test on 2026-11-17, 1024x1024 came back as 1254x1254 and 1792x896 as 1774x887. The other GPT models return the exact size.
Nano Banana works from a ratio and a resolution: the exact pixels are whatever Google returns, at the ratio and resolution of the size you pass.
Limits
Big images take fewer per request: at most 10 at 1K, 4 at 2K and 2 at 4K. More returns a 400 whose param is n.
auto is the default and lets the model pick a size that fits the prompt.
Any other value is rejected with a 400 whose param is size. Which sizes a model accepts is listed in the API reference.
Edit images
Image to image with the Cheapest Imagen API. Send one or more reference images to /v1/images/edits with GPT Image or Nano Banana, as files or as links, and describe the change.
Synchronous and asynchronous requests
Two ways to call the image API. Wait for the image in one request, or send async true, get a task id at once, and read one or many tasks with a single polling request or a webhook.