OpenAI Async Image API

One endpoint for text-to-image and edit. The path is chosen from whether image is present.

Models

Model	Notes
`gpt-image-2`	Default. Quality: `low` / `medium` / `high` / `auto`
`gpt-image-2.5-flare`	Extra quality `xhigh` / `max` (Firefly channel maps those to `high`)
`gpt-image-2.5-sunburst`	Same as flare for quality

How It Works

Create a task

POST /openai/images:

No image → generate
With image → edit
With image + mask → inpaint

Returns { "taskId": "..." }.

Poll

GET /openai/tasks?taskId=<taskId> until completed or failed.

Read the result

result follows response_format: default url (our S3). Set b64_json for base64. Only one field is returned.

image and mask must be http(s):// URLs. Base64 / data URL is rejected here — use Compat generate or Compat edit.

Quick Examples

curl -X POST "https://api.mountsea.ai/openai/images" \
  -H "Authorization: Bearer your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A photorealistic cat wearing a space helmet, floating in orbit",
    "model": "gpt-image-2.5-flare",
    "size": "2048x1152",
    "quality": "high",
    "response_format": "url"
  }'

curl -X POST "https://api.mountsea.ai/openai/images" \
  -H "Authorization: Bearer your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Add a vintage film grain and warm color tone",
    "model": "gpt-image-2",
    "image": "https://example.com/source.jpg",
    "size": "1024x1024",
    "input_fidelity": "high"
  }'

curl -X POST "https://api.mountsea.ai/openai/images" \
  -H "Authorization: Bearer your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Replace the background with a sunset beach",
    "model": "gpt-image-2",
    "image": "https://example.com/source.png",
    "mask": "https://example.com/mask.png"
  }'

curl -X POST "https://api.mountsea.ai/openai/images" \
  -H "Authorization: Bearer your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Wide landscape, golden hour",
    "model": "gpt-image-2",
    "size": "1280x960"
  }'

curl -X GET "https://api.mountsea.ai/openai/tasks?taskId=your-task-id" \
  -H "Authorization: Bearer your-api-key"

Parameters

Field	Type	Required	Description
`prompt`	string	Yes	Prompt (edit: describe the change)
`model`	string	No	Default `gpt-image-2`
`size`	string	No	Preset or custom `WxH` — see below. Default `1024x1024`
`n`	number	No	Count — only `1`
`quality`	string	No	`auto` / `low` / `medium` / `high` / `standard`; 2.5 also `xhigh` / `max`
`background`	string	No	`transparent` / `opaque` / `auto`
`output_format`	string	No	`png` / `jpeg` / `webp`
`output_compression`	number	No	JPEG / WebP 0–100
`moderation`	string	No	`auto` / `low`
`response_format`	string	No	`url` (default, our S3) or `b64_json`. Strictly one field
`input_fidelity`	string	No	`high` / `low` — edit only
`image`	string / string[]	No	`http(s)` URL only. Presence → edit
`mask`	string	No	`http(s)` URL only. Transparent pixels are redrawn

Size

Presets: auto 1024x1024 1536x1024 1024x1536 2048x2048 2048x1152 3840x2160 2160x3840 Custom WxH (e.g. 1280x960) must satisfy all:

Each side ≤ 3840 px
Width and height are multiples of 16
Long : short ≤ 3:1
Total pixels ∈ [655360, 8294400]

Invalid custom size, or a channel that does not support custom size, is rejected.

Task Status

Status	Meaning
`pending`	Queued
`ready`	Ready
`assigned`	Assigned
`processing`	Running
`completed`	Done — read `result`
`failed`	See `errorMessage` / `errorCode`
`cancelled`	Cancelled
`timeout`	Timed out

Terminal results include elapsedMs.

Poll every 2–5 seconds. Typical 10–60 seconds depending on quality and size.

Endpoints

Endpoint	Method	Description
`/openai/images`	POST	Create generate / edit task
`/openai/tasks`	GET	Poll by `taskId`

Sync official SDK? See OpenAI Compat.

​OpenAI Async Image API

​Models

​How It Works

​Quick Examples

​Parameters

​Size

​Task Status

​Endpoints

OpenAI Async Image API

Models

How It Works

Quick Examples

Parameters

Size

Task Status

Endpoints