GPT Image 2

When calling gpt-image-2 through Qiuqiu Token, use this page to prepare image generation and image editing requests. The model name inside Qiuqiu Token remains gpt-image-2. Complete field definitions, model limits, and the latest enum values are subject to the official OpenAI API documentation.

Before starting, prepare balance in Quick Start, then create a grouped token in Create an API Token. Choose a group whose name includes "CodeX dedicated".

Recommended async image generation URL: https://img.qiuqiutoken.com/.

You can also use the online GPT Image 2 generation website directly: https://image.qiuqiutoken.com/.

Generate Images: `POST /v1/images/generations`

Official references

This section only lists fields commonly used with Qiuqiu Token. For full fields, enum values, and response structure, use the OpenAI docs:

Minimal request:

curl

curl https://api.qiuqiutoken.com/v1/images/generations \
  -H "Authorization: Bearer YOUR_IMAGE_TOKEN_KEY" \
  -H "Content-Type: application/json" \
  -d "{
    \"model\": \"gpt-image-2\",
    \"prompt\": \"An orange cat wearing a transparent raincoat on a neon rainy street, cinematic composition\",
    \"size\": \"1024x1024\",
    \"quality\": \"high\",
    \"output_format\": \"png\"
  }"

Common Fields

Field	Required	Notes
`prompt`	Required	Image prompt. Include subject, scene, lighting, composition, and style.
`model`	Recommended	Use `gpt-image-2` through the Qiuqiu Token gateway.
`size`	Optional	Common values include `1024x1024`, `1536x1024`, `1024x1536`, and `auto`.
`quality`	Optional	Common values are `low`, `medium`, `high`, and `auto`. Higher quality is usually slower and more expensive.
`background`	Optional	Common values are `transparent`, `opaque`, and `auto`. Transparent output requires `png` or `webp`.
`output_format`	Optional	Common values are `png`, `jpeg`, and `webp`.
`output_compression`	Optional	Official range is `0` to `100`; meaningful only for `jpeg` or `webp`.
`n`	Optional	Number of images. Official range is `1` to `10`.
`moderation`	Optional	Common values are `low` and `auto`.
`user`	Optional	End-user identifier for audit, risk control, or logs.

Response

For GPT Image models, usually read data[*].b64_json. The most common return value is Base64 image content, not a URL.

{
  "created": 1711111111,
  "data": [
    {
      "b64_json": "iVBORw0KGgoAAAANSUhEUgAA..."
    }
  ]
}

Edit Images: `POST /v1/images/edits`

Upload the source image, optional mask, and prompt with multipart/form-data. The OpenAI reference describes input images as an images array; with multipart upload, image[]=@... is a common form.

curl

curl https://api.qiuqiutoken.com/v1/images/edits \
  -H "Authorization: Bearer YOUR_IMAGE_TOKEN_KEY" \
  -F "model=gpt-image-2" \
  -F "image[]=@./input.png" \
  -F "mask=@./mask.png" \
  -F "prompt=Keep the subject pose, change the background to a neon rainy street, and enhance cinematic reflections" \
  -F "size=1024x1024" \
  -F "quality=high" \
  -F "background=transparent" \
  -F "input_fidelity=high" \
  -F "output_format=png" \
  -F "n=1"

Common Fields

Field	Required	Notes
`images` / `image[]`	Required	Input images. GPT Image officially supports up to 16 input images.
`prompt`	Required	State what to keep, replace, or add.
`model`	Recommended	Use `gpt-image-2` through the Qiuqiu Token gateway.
`mask`	Optional	Transparent areas indicate editable regions.
`size`	Optional	Common values include `1024x1024`, `1536x1024`, `1024x1536`, and `auto`.
`quality`	Optional	Common values are `low`, `medium`, `high`, and `auto`.
`background`	Optional	Use `png` or `webp` for transparent background.
`input_fidelity`	Optional	Common values are `low` and `high`.
`output_format`	Optional	Common values are `png`, `jpeg`, and `webp`.
`n`	Optional	Number of returned images.

Save the Image

When the response contains b64_json, decode it before saving the image file.

python - <<'PY'
import base64

b64 = "PASTE_B64_JSON_HERE"
with open("gpt-image-output.png", "wb") as f:
    f.write(base64.b64decode(b64))
PY

If you use background=transparent, prefer png or webp when saving.

Troubleshooting

401 or authentication failure: use Authorization: Bearer and a valid image group token, not default.
404: confirm the endpoint is /v1/images/generations or /v1/images/edits, and the model is gpt-image-2.
Generation parameter error: return to the minimal JSON example, then add fields back one by one.
Edit file field error: make sure image[] and mask are file uploads or officially supported image references, not plain JSON strings.
Result cannot be opened: save decoded binary image data, not the raw Base64 string.
Transparent background is missing: use background=transparent and output_format=png or webp.