GPT-Image-2 image generation

• Model: gpt-image-2
• Endpoints: sync POST https://apimaster.ai/v1/images/generations · async POST https://apimaster.ai/v1/images/generations/async
• Features: Text-to-image; image-to-image with up to 16 reference images (URL + base64 mix)

Do not call this model via /v1/chat/completions. You will get HTTP 400 with a pointer to the Images API.

Call modes

Mode	Endpoint	When to use
Sync (recommended)	POST https://apimaster.ai/v1/images/generations	OpenAI SDK compatibility — one request returns { "data": [{ "url" }] } (platform polls server-side)
Async submit	POST https://apimaster.ai/v1/images/generations/async	Long jobs, batch pipelines — returns task_id immediately; you poll for the result

Both modes use the same JSON body (model, prompt, size, resolution, etc.).

Quick start (sync)

curl -s "https://apimaster.ai/v1/images/generations" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-image-2","prompt":"An orange cat at sunset, watercolor"}'

Success response

{
  "created": 1717500000,
  "data": [{ "url": "https://apimaster.ai/imgs/....png" }]
}

Client timeout hints (sync)

Tier	Suggested read timeout
1k default	≥ 180s
2k / medium	≥ 300s
4k / high	≥ 600s

Auth

Authorization: Bearer YOUR_API_KEY
Content-Type: application/json

Body parameters

Field	Type	Required	Notes
model	string	yes	gpt-image-2
prompt	string	yes	Scene description; safety moderation applies
n	integer	no	1 only (numeric, no quotes)
size	string	no	Aspect ratio, default 1:1
resolution	string	no	1k / 2k / 4k, default 1k
image_urls	array	no	Reference images → image-to-image
official_fallback	boolean	no	Allow upgrade to official-capable channel on standard-request retry, default false

size values

auto, 1:1, 3:2, 2:3, 4:3, 3:4, 5:4, 4:5, 16:9, 9:16, 2:1, 1:2, 3:1, 1:3, 21:9, 9:21, or pixel strings like 3840x2160.

resolution (sample pixels)

size	1k	2k	4k
1:1	1024×1024	2048×2048	2880×2880
16:9	1536×864	2048×1152	3840×2160

All 15 ratios support 4K.

image_urls

• Max 16 images
• Public URLs and base64 data URIs can be mixed
• Omit size to follow input resolution; set size to force output aspect ratio

Examples

16:9 @ 2K (sync)

curl -s "https://apimaster.ai/v1/images/generations" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "a corgi astronaut on the moon, cinematic",
    "size": "16:9",
    "resolution": "2k"
  }'

Errors (sync)

HTTP	Meaning
400	Bad parameters, or wrong endpoint (chat)
401	Bad API key
402	Insufficient balance
408	Generation timeout — lower resolution/quality or switch to async
429	Rate limited

Async mode: submit

POST https://apimaster.ai/v1/images/generations/async with the same body. Returns immediately:

{
  "code": 200,
  "data": [{ "status": "submitted", "task_id": "task_01KPQ7J7DWB7QZ3WCEK3YVPBRA" }]
}

Async mode: poll result

curl -s "https://apimaster.ai/v1/tasks/TASK_ID?model=gpt-image-2" \
  -H "Authorization: Bearer YOUR_API_KEY"

• ?model=gpt-image-2 selects the correct channel
• Image URL when done: data.result.images[0].url[0]
• Submit response (/images/generations/async) uses status: submitted. Poll response (GET /tasks/...) often returns pending while generating; processing / in_progress are also possible. Terminal states: completed or failed.

Phase	Typical status	Meaning
Submit OK	submitted	Only in async submit data[]
Queued / running	pending, processing, in_progress	Keep polling — pending is normal, not an error
Success	completed	Read data.result.images[0].url[0]
Failure	failed, error, cancelled	Inspect error fields in the response

• Poll after 10–20s, every 3–5s

Notes

• Use Images API endpoints, not Chat Completions.
• Prefer size for aspect ratio; avoid repeating ratio in prompt.
• Billing by resolution tier; see console for your account.