GPT Image 2.5

Use an Official Group API key. Flare and Sunburst have distinct public model IDs; these are standalone image requests, not Responses tool calls on a GPT text model.

Supported protocol

Generate with JSON POST /v1/images/generations, or upload files with POST /v1/images/edits (multipart/form-data). Submission returns 202 Accepted and a task id. Poll GET /v1/tasks/{id} until completed or failed. A 202 response is not a completed image.

This family uses Images + asynchronous tasks. Image generation through /v1/chat/completions or /v1/responses is not supported. See the complete submission and polling example.

Models and parameters

size selects an aspect ratio such as 1:1 or 16:9; resolution selects the resolution tier. quality accepts auto, low, medium, high, xhigh and max. The example uses low. Prompts are limited to 4000 characters. Substitute gpt-image-2.5-sunburst to use Sunburst.

Generate an image

Set APIPOOL_API_KEY to your key, then run:

APIPOOL_ENDPOINT="https://app.apipool.dev"

curl "$APIPOOL_ENDPOINT/v1/images/generations" \
  -H "Authorization: Bearer $APIPOOL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.5-flare",
    "prompt": "A blue ceramic cup on a white table",
    "size": "1:1",
    "resolution": "1k",
    "n": 1,
    "quality": "low"
}'

Edit an image

Upload a local PNG, JPEG or WebP file, replacing source.png with its path:

curl "https://app.apipool.dev/v1/images/edits" \
  -H "Authorization: Bearer $APIPOOL_API_KEY" \
  -F "model=gpt-image-2.5-flare" \
  -F "prompt=Change the cup to green" \
  -F "image=@./source.png" \
  -F "size=1:1" \
  -F "resolution=1k" \
  -F "n=1"

For multiple references, repeat -F "image=@./second.png" within the model's limit. To use publicly downloadable references, send image_urls to the same JSON generation endpoint:

{
  "model": "gpt-image-2.5-flare",
  "prompt": "Change the cup to green",
  "size": "1:1",
  "resolution": "1k",
  "n": 1,
  "quality": "low",
  "image_urls": ["https://your-cdn.example/source.png"]
}

Replace the example URL with a real address reachable by the server. Expired tunnels, DNS failures and download timeouts can fail the task. Browser-only localhost or authenticated image URLs are unsuitable.

Poll and retrieve results

Save the submission's id, such as imgtask_01..., and query with the same key:

TASK_ID="imgtask_01..."
curl "https://app.apipool.dev/v1/tasks/$TASK_ID" \
  -H "Authorization: Bearer $APIPOOL_API_KEY"

Continue polling for submitted, submission_unknown, processing or meter_pending. On completed, read data[].url and download before expiry. On failed, inspect error. Do not resubmit while waiting. See Asynchronous image tasks for all states, timeout handling and a runnable polling example.

Scope and billing

APIPool delivers images as asynchronous tasks. Do not assume the synchronous Images response shape of a standard OpenAI SDK call: read the task ID and poll it.

See the selected model and group in Models & pricing. Pricing and available parameter tiers depend on that listing. These models use token pricing and return normalized usage when complete.