
GPT Image 2.5 Flare Edit API by OpenAI
GPT Image 2.5 Flare Edit applies natural-language instructions to up to 16 reference images, with an optional mask, arbitrary resolutions up to 3840x2160, five quality tiers including xhigh and max, and transparent-background output. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.
GPT Image 2.5 Flare Edit è sviluppato da OpenAI. Atlas Cloud (gestita da Atlas Cloud AI LLC) ne fornisce l’accesso ma non ne è proprietaria. Tutti i marchi appartengono ai rispettivi titolari.
GPT Image 2.5 Flare Edit
GPT Image 2.5 Flare is OpenAI's upgraded GPT Image generation model, reached through the OpenAI Images API. Generation is asynchronous: submit the request, then poll for the final image URL.
Highlights
- Up to 16 reference images in a single request, with an optional
mask. - Five quality tiers:
low,medium,high, plusxhighandmax, which are new in gpt-image-2.5. - Arbitrary resolutions up to 3840x2160, and fully supported transparent backgrounds.
- Long prompts: up to 32000 characters.
Parameters
| Parameter | Required | Description |
|---|---|---|
model | Yes | openai/gpt-image-2.5-flare/edit |
images | Yes | Up to 16 source images. Each entry is a URL, Base64 data, or an asset reference (asset://<ASSET_ID>). |
mask | No | Optional mask marking the region to edit, in the same formats as images. Transparent pixels mark the area the model should replace; it must match the first source image's dimensions. |
prompt | Yes | Instruction describing the desired edit. Up to 32000 characters. |
quality | No | Rendering tier: low, medium, high, xhigh, max. Defaults to medium. xhigh and max are new in gpt-image-2.5. |
size | No | Output size as WIDTHxHEIGHT (for example 1536x864) or auto. Defaults to 1024x1024. The only control over output dimensions. |
background | No | auto, opaque, or transparent. transparent forces a png or webp output format. |
output_format | No | png, jpeg, or webp. Defaults to png. |
n | No | Number of images to generate, 1 to 10. Defaults to 1. Each image is billed separately. |
How To Use
curl -X POST "https://api.atlascloud.ai/api/v1/model/generateImage" \ -H "Authorization: Bearer $AIP_API_KEY" \ -H "Content-Type: application/json" \ --data-raw '{ "model": "openai/gpt-image-2.5-flare/edit", "images": [ "https://static.atlascloud.ai/media/images/example-a.png", "https://static.atlascloud.ai/media/images/example-b.png" ], "prompt": "Replace the background with a sunlit meadow, keep the subject unchanged.", "quality": "high", "size": "1536x1024" }'
Size
size is the only control over output dimensions. gpt-image-2.5 accepts
arbitrary resolutions, subject to:
- both edges at least 256 px, floored to a multiple of 16
- each edge at most 3840 px, and at most 3840x2160 in total
- aspect ratio between 1:3 and 3:1
Resolutions above 2560x1440 are experimental upstream. auto lets the model
choose, and is billed at the auto ceiling then settled down against the image
actually produced.
Notes
-
Generation is asynchronous. Submit the request, then poll
/api/v1/model/prediction/{request_id}for the final image URL. -
Pricing is per token, at OpenAI's list rates: image tokens 30.00/1M out, text input 2.00/1M and $1.25/1M respectively). The request is charged from the token usage the model actually reports, so the amount held at submit is an upper bound and the difference is refunded.
-
Streaming (
stream,partial_images) andresponse_formatare not supported — the API always returns finished images re-hosted on AtlasCloud storage. -
Source images and the mask are fetched and uploaded to the upstream API as multipart files, so URLs, Base64 payloads and
asset://references are all accepted.



















