GPT Image 2.5

Request and parameters

{
  "model": "gpt-image-2.5",
  "variant": "sunburst",
  "mode": "image_generation",
  "input": { "prompt": "A product photograph on a blue background", "images": [] },
  "parameters": { "size": "1024x1024", "quality": "medium", "output_format": "png", "background": "auto", "num_outputs": 1 }
}

One route, two variants: sunburst (default) and flare. No input images selects generation; images select editing. Supply up to 4 HTTPS URL or media_asset locators in input.images. Prompt: 1–4000 characters. num_outputs: integer from 1 to 10, default 1. All results are returned in artifacts.

Default size: 1024x1024. Custom WIDTHxHEIGHT: multiples of 16, edges up to 3840, 655360–8294400 pixels, aspect up to 3:1. Quality: low, medium (default), high, xhigh, max. Format: png (default) or jpeg. Background: auto (default), opaque or transparent; transparency requires PNG. Gensta does not yet expose size/quality=auto, WebP or masks.

Final billing uses actual text, image, cached input and output tokens. GET /v1/models exposes current customer rates. An estimated amount with headroom is reserved before submission; unused funds are released. Ambiguous submissions never automatically repeat a paid generation.