GPT Image 2.5 Flare

Q: How do I get an API key?

Sign in, open API keys and create a key. Keep it on your server and never include it in public client code.

Model IDgpt-image-2.5-flare

Generate or edit images using a prompt and optional references.

Interface

OpenAI
Image
Asynchronous task API

From $5.60 / 1 million tokensInput

View price table

Get an API key

Integration guide (Markdown)Model parameters (JSON)

Input

Prompt

0 / 32,000

Enter to generate · Shift+Enter for a new line

Reference imagesOptional

0 / 16

Mask imageOptional

0 / 1

Image count

Quality

Image size

Advanced settings

Background

Image objects

Mask

File ID

Reference imageOptional

0 / 1

Moderation

Output compression

Output format

Response format

User identifier0 / 512

Result

Idle

Run a request to preview the result here.

Use a prompt and supported parameters. Submission starts an asynchronous task.

GPT Image 2.5 Flare API integration guide

Generate or edit images using a prompt and optional references. Use gpt-image-2.5-flare with the asynchronous task API: submit once, save task_id and poll the result. An accepted submission is not a completed generation.

Parameters reviewed on 2026-10-11

Before you integrate

Model ID: gpt-image-2.5-flare
Output type: Image
API protocol: Asynchronous task API
Resolution: Not published
Aspect ratio: Not published
Reference images: Supported
Artifact retention: 14 days; download before expiry
Base URL: https://open.scenepond.ai/v1

Prepare only the media and parameters accepted by this interface. Accessible reference URLs are required for API requests.

Strengths and limits

Set subject, composition and lighting explicitly. Compare detail and consistency before accepting the image.
For GPT Image, compare supported quality settings using the same prompt; token usage determines the final image bill.

The published parameters describe this interface. Unlisted capabilities have not been verified. Review output quality and input rights before publication.

Price table

Specification	Unit price	Billing unit
Standard · Reference images	$5.60	1M tokens
Standard · Image output	$21.00	1M tokens
Standard · Text input	$3.50	1M tokens

Example bill: 1 image · Illustrative token estimate $0.0224

Prices depend on model settings. Usage records show recorded deductions; asynchronous task charges may be adjusted after completion.

Your first request

Create an API key and set SCENEPOND_API_KEY in your server environment. Never expose the key in browser code.

curl --fail-with-body -X POST "https://open.scenepond.ai/v1/tasks" \
  -H "Authorization: Bearer $SCENEPOND_API_KEY" \
  -H "Content-Type: application/json" \
  --data-binary @- <<'JSON'
{
  "model": "gpt-image-2.5-flare",
  "parameters": {
    "count": 1,
    "prompt": "A quiet lakeside at sunrise, soft light and a clear subject."
  }
}
JSON

Confirm the result

Save task_id from the creation response. Status and progress are a receipt, not the finished media.
GET /v1/tasks/{task_id} — Poll every few seconds until succeeded or failed. Do not create another task to check progress.
GET /v1/tasks/{task_id}/artifacts — After succeeded, list artifacts and download artifacts[].content_url with the same API key before artifacts[].expires_at.
On failure, read error.code and error.message; inspect error.retryable and error.charged when present. If absent, check usage and ask support before retrying.

curl --fail-with-body -X GET "https://open.scenepond.ai/v1/tasks/$SCENEPOND_TASK_ID" \
  -H "Authorization: Bearer $SCENEPOND_API_KEY"

curl --fail-with-body -X GET "https://open.scenepond.ai/v1/tasks/$SCENEPOND_TASK_ID/artifacts" \
  -H "Authorization: Bearer $SCENEPOND_API_KEY"

Troubleshoot by symptom

HTTP errors describe request failures. A task can also fail after a successful HTTP response; read its status and error fields.

HTTP status	Error code	Cause and action
409	`reservation_changed`	The reservation has changed. Check the latest estimate and confirm again. Refresh the reservation estimate, review the amount, then submit explicitly.
401	`invalid_api_key`	The API key is invalid or expired. Create a new key and check the Authorization header. Keep the key on your server. Replace revoked keys and send Authorization: Bearer.
410	`artifact_expired`	The result has expired. Use a downloaded copy; expired files cannot be recovered. Download successful artifacts before expires_at. Retrying a download cannot restore an expired file.
200 · failed	`content_policy_violation / content_moderation`	Content was rejected by the safety policy. Revise the prompt or reference media. Revise the prompt and references. Do not retry the same rejected content.
400 / 422	`invalid_parameters / invalid_request`	Check required parameters, types, and allowed values. Compare your input with the parameter schema, including required media and allowed values.
200 · failed	`timeout / task_timeout`	The task timed out. Check its status and usage before resubmitting. Query the existing task and check usage. A timed-out submission may have been accepted.
200 · failed	`model_unavailable`	This model is temporarily unavailable. Try another enabled model. Wait or select a related model. Check charged and retryable before submitting again.
402	`insufficient_balance`	Your balance is insufficient. Add credit before submitting. Add credit, then submit a new task after checking the previous attempt.
429	`rate_limit_exceeded`	Respect the returned retry delay and lower your request rate. Follow Retry-After when present and use bounded backoff. Lower concurrency.
200 · failed	`generation_failed`	Generation failed. A detailed cause was not returned; check usage before resubmitting. Preserve the task ID and contact support if the returned error does not explain the failure.

When contacting support, include the time, model ID, task ID and error information. Never send your API key.

Use cases and prompt examples

Product photograph

Prepare a product image with controlled lighting.

A matte ceramic cup on a pale wooden table, soft window light from the left, neutral background, realistic glaze texture, clean composition, no text or watermark.

Editorial illustration

Explore a visual concept for an article.

An editorial illustration of a small greenhouse in a busy city, plants growing around the glass walls, warm green and amber palette, clear focal point, no lettering.

Composition study

Compare framing for a campaign visual.

A red bicycle leaning against a light stone wall, afternoon shadows, generous empty space on the right, natural proportions, understated photographic style, no text.

Frequently asked questions

How is usage billed?

Calls are billed by actual usage at the published rates. Parameter choices can change the bill; final deductions are shown in usage records.

Are failed tasks charged?

Read error.charged when returned and check usage records. A false value means no charge; missing billing information is not a promise of a free retry.

How long are results retained?

Generated artifacts are stored for 14 days. Download them before expires_at; expired artifacts cannot be recovered.

Can I use the output commercially?

Review the user agreement and confirm that you have the rights to your inputs and intended use.

User agreement

What are the concurrency and rate limits?

Limits depend on your account and API key. Follow Retry-After when present, lower concurrency and use bounded backoff.

How do I get an API key?