GPT Image 2 All API reference

Authentication

All requests carry the API key in the header:

Authorization: Bearer YOUR_API_KEY

Pricing (resolution × quality)

Billed per image; quality=auto is charged at the medium tier; failed requests are not charged. Speed scales with quality: low ~40s, medium ~60-90s, high ~2 minutes.

Sizes & billing tiers

Arbitrary WxH is also supported: both sides multiples of 16, long edge ≤ 3840, aspect ratio ≤ 3:1, total pixels 655K-8.29M. Alternatively omit size and pass aspect_ratio + resolution (1K/2K/4K) and the server derives the best size.

Output sizes: 2K and 4K

The 2K and 4K tiers follow OpenAI's own size labels for gpt-image-2. These are the 2K and 4K sizes listed in OpenAI's image generation guide:

• The billing tier follows the size you request, not the size delivered. A 2K request such as 2048x1152 can return exactly 2048x1152, OpenAI's 2K landscape size, or a larger 2736x1536 at the same ratio; both are billed as 2K.
• The same guide notes that resolutions above 2560x1440 are experimental, so a larger size such as 2736x1536 isn't guaranteed on every request.
• 4K is capped by OpenAI's limit of 8,294,400 total pixels (3840x2160). 16:9 and 9:16 reach 3840 on the long edge; other ratios have a shorter long edge, for example 2880x2880 at 1:1.

Source: OpenAI image generation guide · GPT Image 2 settings

Option 1: OpenAI SDK / Codex / Cursor (sync)

Passing an OpenAI-style WxH size triggers sync mode: the response carries the image inline (b64_json + url), matching the official images.generate() / images.edit() contract.

In sync mode the top-level size is the actual size of the image we delivered (since 2026-10-25), not an echo of your request, and each data item also carries width and height. If the image could not be measured, size behaves as it did before.

# Text-to-image — OpenAI Images API shape, SYNC response (no task polling)
curl -X POST https://api.apimodels.app/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2-all",
    "prompt": "a white sneaker product shot on beige background, studio light",
    "size": "2048x2048",
    "quality": "medium",
    "n": 1
  }'
# → { "created": ..., "data": [{ "b64_json": "...", "url": "https://r2.apimodels.app/...", "width": 2048, "height": 2048 }], "size": "2048x2048" }

# Image edit / multi-image fusion — multipart, OpenAI images.edit() shape
curl -X POST https://api.apimodels.app/v1/images/edits \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "model=gpt-image-2-all" \
  -F "image=@product.png" \
  -F "prompt=put this sneaker on a model, e-commerce close-up" \
  -F "size=2048x2048" \
  -F "quality=medium"

Option 2: native async (taskId + poll / callback)

Without a WxH size (or when you need callback_url) the request follows the platform async contract: create returns a taskId; poll until completed, or receive a webhook callback.

# Native async contract — pass aspect_ratio + resolution tier instead of size.
curl -X POST https://api.apimodels.app/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2-all",
    "prompt": "a lighthouse on a cliff at dusk",
    "aspect_ratio": "9:16",
    "resolution": "2K",
    "quality": "medium",
    "callback_url": "https://your-domain.com/webhook"
  }'
# → { "code": 200, "data": { "taskId": "..." } }

curl "https://api.apimodels.app/v1/images/generations?task_id=TASK_ID" \
  -H "Authorization: Bearer YOUR_API_KEY"

Parameters

Notes

• Same underlying model as gpt-image-2 (the async task-id channel); this channel adds OpenAI compatibility, sync responses and the full quality matrix.
• Upload size: requests through apimodels.app are capped at ~4.5MB per request body (413 beyond that). For large images or many reference images, switch base_url to https://api.apimodels.app/v1 — same API, same key, 100MB per request.
• Speed scales with quality: low ~40s, medium ~60-90s, high ~2 minutes. For sync calls set client timeouts to ≥180s.
• Failed generations are not charged. Result files are auto-deleted after 30 days — download or re-host promptly.

Try it in the Playground