AI Image Generation

Text to image

Describe the image in prompt and choose an image model. Generated images are PNG files.

{
    "asset": {
        "type": "image",
        "prompt": "Aerial photograph of Sydney Harbour at dusk, city lights reflecting on the water",
        "model": "nano-banana-2",
        "options": {
            "aspectRatio": "16:9",
            "resolution": "2K"
        }
    },
    "start": 0,
    "length": 5
}

Editing images

The edit models take one or more existing images in options.imageUrls and change them as the prompt describes. Use them to retouch a product shot, restyle a photo or combine several images into one. Say what to change and what to keep.

{
    "asset": {
        "type": "image",
        "prompt": "Place the handbag from the first image on the marble bench from the second image. Keep the handbag's colour and stitching exactly as they are.",
        "model": "nano-banana-2-edit",
        "options": {
            "imageUrls": [
                "https://example.com/images/handbag.jpg",
                "https://example.com/images/marble-bench.jpg"
            ],
            "aspectRatio": "16:9"
        }
    },
    "start": 0,
    "length": 5
}

imageUrls must be public HTTP or HTTPS URLs. nano-banana-2-edit takes up to 14 images and gpt-image-2.5-sunburst-edit takes up to 16, in order.

Models

Model	Use it for	Resolution
nano-banana-2	The default. Fast, with accurate text and coherent compositions.	0.5K, 1K (default), 2K, 4K
nano-banana-2-edit	Editing and combining images with Nano Banana 2.	0.5K, 1K (default), 2K, 4K
nano-banana-pro	Dense compositions with many instructions, at high resolution.	1K (default), 2K, 4K
gpt-image-2.5-sunburst	Highly detailed images that follow complex prompts closely. Slower.	1K (default), 2K, 4K
gpt-image-2.5-sunburst-edit	Precise edits with up to 16 references. Slower.	1K (default), 2K, 4K
flux-schnell	Quick illustrations and drafts. Takes no options.	Fixed

Options

Option	Models	Values
aspectRatio	Nano Banana 2 and its edit model	auto (default), 21:9, 16:9, 3:2, 4:3, 5:4, 1:1, 4:5, 3:4, 2:3, 9:16, 4:1, 1:4, 8:1, 1:8
	Nano Banana Pro	auto, 21:9, 16:9, 3:2, 4:3, 5:4, 1:1 (default), 4:5, 3:4, 2:3, 9:16
	GPT Image models	21:9, 16:9 (default), 3:2, 4:3, 5:4, 1:1, 4:5, 3:4, 2:3, 9:16
resolution	All except flux-schnell	See the table above.
imageUrls	Edit models	The images to edit. Required.
seed	Nano Banana models	An integer. The same seed and prompt reproduce a result.
systemPrompt	Nano Banana models	Standing instructions sent with the prompt, up to 2,000 characters.

GET /models/{id} returns the full option schema for a model.

Writing image prompts

This guidance is for the Nano Banana models.

• Write plain sentences. Describe the scene as you would to a photographer. Shot type, camera angle, lighting, lens and focal length all work. Comma-separated tags and quality boosters like "8k" don't help.
• There is no negative prompt. Say what you want instead: "an empty desk", not "no clutter".
• Quote text that must appear, and describe the font in plain words: the text "Grand Opening" in a clean, bold, sans-serif.
• Use Pro for hard compositions, not for better images in general. Nano Banana 2 matches or beats Pro on most images at about half the price.

For copy that must be spelled exactly, run to several lines or change with merge fields, generate a plain background and put a rich text asset over it. Real fonts are always spelled correctly, stay sharp at any size and can be reworded without generating again.

Match aspectRatio to the clip it fills. An image that will become the start of a generated video should be larger than the video: use 2K for a 720p or 1080p video.