GPT Image 2 Image Generation

##All APIs require Bearer Token authentication## **Get API Key:** Visit [API Key Management Page](https://starmagic.ai/app/api-keys) to get your API Key **Add to request header:** ``` Authorization: Bearer YOUR_API_KEY ```

Authorization: Bearer YOUR_API_KEY

Image generation model name, official channel, better stability and controllability, suitable for commercial scenarios

"gpt-image-2"

Prompt describing the image to be generated, or describing how to edit the input image **Limits:** - Up to `32000` characters (counted by Unicode code points, works for CJK and other languages)

"A beautiful colorful sunset over the ocean"

Reference image URL list for image-to-image and image editing functions **Note:** - Number of input images per request: `1~16` - Size of a single image: not exceeding `50MB` - Supported file formats: `.jpeg`, `.jpg`, `.png`, `.webp` - Image URLs must be directly accessible by the server, or the image URL should directly download when accessed (typically these URLs end with image file extensions, such as `.png`, `.jpg`) - In image-to-image / image editing scenarios, the reference images themselves also incur additional image input token consumption

[
  "https://example.com/image1.png",
  "https://example.com/image2.png"
]

Inpainting mask URL — marks the region of the reference image to regenerate. **Only valid in image edit mode** (must be combined with `image_urls`); the mask is silently ignored in pure text-to-image requests. **Format requirements:** - **Must be a PNG with an alpha channel**: transparent pixels (`alpha < 255`) = regions to regenerate; opaque pixels = preserved - **Mask dimensions must exactly match the reference image dimensions** (width × height in pixels) - Single mask per request **Note:** - At least one reference image is required in `image_urls`; a mask sent alone has no effect - Common errors: - `Invalid mask image format - mask image missing alpha channel`: the uploaded image has no alpha channel (JPEG, opaque PNG, etc.). Re-export the mask as a PNG with transparent regions. - `Invalid mask image format - mask size does not match image size`: the mask dimensions don't match the reference image. Resize the mask to the same pixel dimensions as your reference image.

"https://example.com/mask.png"

Size of the generated image. Supports both **ratio format** and **explicit pixel format**, defaults to `auto` **① Ratio format (recommended, 15 options)** - `1:1`: Square - `1:2` / `2:1`: Extreme portrait / landscape - `1:3` / `3:1`: Ultra portrait / landscape (3:1 limit) - `2:3` / `3:2`: Standard portrait / landscape - `3:4` / `4:3`: Classic portrait / landscape - `4:5` / `5:4`: Common social media - `9:16` / `16:9`: Mobile / desktop widescreen - `9:21` / `21:9`: Ultra-wide **② Explicit pixel format**: `WxH` (or `W×H`), e.g. `1024x1024`, `1536x1024`, `3840×2160` - Both width and height must be multiples of `16` - Each edge range: `[16, 3840]` - Pixel budget: `655,360 ≤ width × height ≤ 8,294,400` (about 0.65 MP ~ 8.29 MP) - Aspect ratio: `≤ 3:1` **③ `auto`**: The model decides the size automatically (`resolution` does not apply in this mode) **Out-of-range handling:** - If a ratio + `resolution` combination exceeds the pixel budget, dimensions are automatically scaled down proportionally (e.g. 4K 2:1 → 3840×1920)

"auto"

Resolution tier shortcut, only effective when `size` is a ratio; ignored in explicit pixel mode **Pixel budget rules** (dimensions are derived from the target pixel count and the `size` ratio, aligned to multiples of 16): - `1K`: ~1 MP (1024² = 1,048,576 pixels) - `2K`: ~4 MP (2048² = 4,194,304 pixels) - `4K`: ~8.29 MP (3840×2160 = 8,294,400 pixels, the maximum) **Landscape / square output dimensions** (portrait dimensions are the landscape width/height swapped, e.g. `2:3` = `3:2` reversed): | Ratio | 1K | 2K | 4K | |---|---|---|---| | `1:1` | 1024×1024 | 2048×2048 | 2880×2880 | | `2:1` | 1456×720 | 2896×1456 | 3840×1920 \* | | `3:1` | 1776×592 | 3552×1184 | 3840×1280 \* | | `3:2` | 1248×832 | 2512×1680 | 3520×2352 | | `4:3` | 1184×880 | 2368×1776 | 3312×2480 \* | | `5:4` | 1152×912 | 2288×1824 | 3216×2576 | | `16:9` | 1360×768 | 2736×1536 | 3840×2160 (UHD) | | `21:9` | 1568×672 | 3136×1344 | 3840×1632 \* | \* Marks combinations that are auto-downscaled to fit the pixel budget. Values are case-insensitive.

"1K"

Rendering quality that controls the model's "reasoning depth", directly affecting output token count and cost. Defaults to `medium` | Value | Tile base | Relative cost (1024²) | |---|---|---| | `low` | 16 | ~0.11× | | `medium` | 48 | 1.0× | | `high` | 96 | ~4.0× |

"medium"

Number of images to generate, each billed independently **Note:** - Text input tokens scale linearly with `n`

1

HTTPS callback address after task completion **Callback Timing:** - Triggered when task is completed, failed, or cancelled - Sent after billing confirmation is completed **Security Restrictions:** - Only HTTPS protocol is supported - Callback to internal IP addresses is prohibited (127.0.0.1, 10.x.x.x, 172.16-31.x.x, 192.168.x.x, etc.) - URL length must not exceed `2048` characters **Callback Mechanism:** - Timeout: `10` seconds - Maximum `3` retries on failure (retries after `1` second/`2` seconds/`4` seconds) - Callback response body format is consistent with the task query API response format - Callback address returning 2xx status code is considered successful, other status codes will trigger retry

"https://your-domain.com/webhooks/image-task-completed"