Edit an image (OpenAI-compatible)

Server URLloading...

POST

/v1/images/edits

Send

Authorization

Body

Accepts multipart/form-data (with a reference image) or JSON. Billed per call.

For openai/gpt-image-2.5-sunburst 2K / 4K edits, send up to 10 reference images as multipart image / image[], or as JSON image / images (https URLs or data URLs; images:[{image_url}] is also accepted).

Per-model limits (openai/gpt-image-2.5-sunburst, openai/gpt-image-2.5-flare, openai/gpt-image-2):

• Standard size: auto (default), 1024x1024, 1536x1024, 1024x1536. Output is about 1.5 MP; size only sets the aspect ratio and quality does not change the output or the price. Other sizes return 400.
• n: up to 4 for openai/gpt-image-2; 1 for flare and sunburst.
• The response is always {created, data:[{b64_json}]} (base64 PNG); response_format has no effect.
• openai/gpt-image-2.5-sunburst 2K / 4K: 2048x2048, 2560x1440, 1440x2560 or "2K" = 2K at $0.08 per image; 3840x2160, 2160x3840 or "4K" = 4K at $0.12 per image (standard is $0.0186). 2K / 4K requires n = 1, ignores quality, does not support mask, and is never retried or served by a backup model. Unlisted sizes return 400 and are not charged.

xai/grok-imagine-image-2.0: send one source image as JSON image ({"url": ...}, an https URL or a data URL), up to 3 as JSON images, or as multipart image / image[]. With several images, refer to them in the prompt as <IMAGE_0>, <IMAGE_1> and so on. For edits quality defaults to medium (auto means medium); resolution and n work as for generations, and there is no mask. Billed per output image by quality × resolution, plus each source image once per request. More than 3 source images return 400 before anything is billed.

See Images: Parameters vs. OpenAI for the full list of supported and unsupported parameters.

Response Body

curl -X POST "https://example.com/v1/images/edits" \  -H "Content-Type: application/json" \  -d '{}'

{  "model": "string",  "stream": true}