
Nano Banana
Light image gen. Up to 3 reference images. Good for testing composition.
Try one first. Up to 3 reference images.
Beta uses fixed settings and one image per request. Use it to try composition, style and prompts before opening the full quality tiers.
Beta uses fixed settings and one image per request. Use it to try composition, style and prompts before opening the full quality tiers.
Just launched
Nano Banana, Seedream, Grok, Qwen, and GPT Image are all here. Open a card for that model's page.

Light image gen. Up to 3 reference images. Good for testing composition.
Try one first. Up to 3 reference images.

Second banana line. Up to 14 reference images. 1K/2K/4K.
Need more reference images? Switch to 2.

A cleaner 2. Same 14 reference images, cleaner light.
Need cleaner studio light? Switch to 2.1.

A lighter 2. Up to 14 reference images. Resolution stays at 1K.
Need 2's references without 4K? Use Lite.

The professional banana line. Up to 14 reference images. For delivery.
Need the full banana line? Use Pro.

Scene-first. Up to 10 reference images. 1K/2K. Grows a full frame from one point.
Need a scene that grows? Switch to Seedream.

xAI image gen. Up to 5 reference images. Harder contrast and colder light.
Need hard contrast and deep space? Switch to Grok.

Qwen image gen. Up to 10 reference images. Ink-wash, Chinese style, and modern light.
Need ink-wash plus modern light? Switch to Qwen.

Low-cost start. Fixed settings, one image per request. Good for testing style and composition.
Try the look first. Lower price.

Official API. Quality, size and count are all adjustable. For production images.
Need control and stability? This one.

2.5 model. Sharper flare and harder edges. For night, metal and neon.
Need hard light and contrast? Switch to Flare.

2.5 model. Stronger sunburst and volume light. For sunrise and backlit portraits.
Need sun and backlight? Switch to Sunburst.
Specs
No quality picker, no resolution picker, no batch. Put the work into the prompt.
For lots of trials. Stay on Beta while the prompt is still moving.
No quality or 2K/4K picker. Send it and get an image.
When the composition is right, switch to Image 2 or 2.5 for quality and count.
Into your workflow
Generate in the dashboard, or take the same Model ID into the API.
Pick this model in the dashboard, write a prompt, and generate.
OpenAI-compatible API. Paste the exact Model ID.
Model ID: gpt-image-2-beta