GPT Image 2
Generate or edit images from text and up to sixteen public HTTPS reference images, with output up to 4K.
Supported modes
Async flow: Create the task, save data.id, then poll the Task with GET /v1/tasks/{task_id}. On succeeded, read data.output.media. On failed, read data.error_code and data.error_message, then follow the error and retry policy.
Using a local file? upload a local input first and pass the returned data.url.
認証
Send your API key as Authorization: Bearer <BEATAPI_API_KEY>.
ヘッダー
リクエスト
Must be gpt-image-2.
Generation or image-editing instructions.
Public HTTPS reference-image URLs in PNG, JPEG or WebP; each link is checked when the task is created, and one that is not such an image (an error page, a GIF, a 404) is refused with 400. Omit for text-to-image.
Output image aspect ratio. Send resolution and aspect_ratio, or size instead of both — never together.
Output resolution tier. Not with size.
Exact output size as WIDTHxHEIGHT, for example 1536x1024 or 3840x2160. Replaces resolution and aspect_ratio; sending either with it is a 400. Both edges must be multiples of 16 and between 16 and 3840 px, the total between 655,360 and 8,294,400 pixels, and the long edge at most three times the short one. Billed as 1K up to 2,097,152 pixels, 2K up to 5,898,240 pixels, and 4K above.
Output background. transparent returns a PNG with an alpha channel — describe the subject as isolated on a transparent background in the prompt. opaque always fills the backdrop; auto lets the model choose.