MiniMax H3 Image

​

Models

• minimax-h3-image — Text-to-image when image_urls is omitted; image editing with 1 to 16 reference images when image_urls is provided

​

Features

• Text-to-image generation from prompts of up to 5,000 characters.
• On-image text rendering for titles, labels, and captions written in quotes in the prompt.
• Image editing and multi-reference composition with up to 16 reference images, referenced in the prompt as <Picture 1>, <Picture 2>, and so on.
• Ten aspect ratios at 1K, 2K, or 4K (up to 6272×2688 at 21:9), plus custom WIDTHxHEIGHT sizes from 655,360 to 20,000,000 pixels.
• Prompt expansion that rewrites a short prompt in any language into a detailed description of the whole image.
• PNG or JPEG output.

​

Input modes

• Text-to-image — Omit image_urls. size defaults to 1:1, and enable_prompt_expansion controls whether the prompt is expanded.
• Image editing — Pass 1 to 16 URLs in image_urls. The first URL is <Picture 1>, the second is <Picture 2>, and so on; use these tags in the prompt to say which image to change and which images to take content from. size defaults to auto, which follows the aspect ratio of <Picture 1>. Image editing requests always expand the prompt.

​

Required parameters

• prompt (string): Description of the image or the requested edit, 1–5,000 characters. Must contain non-whitespace text. Put exact on-image text in quotes and state where it goes.

​

Optional parameters

• image_urls (array of strings): 1 to 16 publicly accessible HTTP(S) reference image URLs (JPEG, PNG, or WEBP, up to 20 MB each, 128 to 8192 pixels per side). Selects image editing mode. Refer to the images in the prompt by their order: <Picture 1> for the first URL, <Picture 2> for the second, and so on.
• size (string): 1:1, 5:4, 4:5, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9, auto, or a custom WIDTHxHEIGHT. Defaults to 1:1 for text-to-image and auto for image editing. auto applies to image editing. See Size.
• resolution (string): 1K, 2K, or 4K. Default is 2K. Sets the pixel tier for aspect-ratio sizes and auto. A custom WIDTHxHEIGHT size sets the exact output pixels.
• enable_prompt_expansion (boolean): Expand the prompt into a detailed description of the whole image before generating. Default is true. Applies to text-to-image requests. Image editing requests always expand the prompt.
• output_format (string): png or jpeg. Default is png.

​

Size

• auto (image editing): The output follows the aspect ratio of <Picture 1> at about the pixel area of the selected tier (1024×1024, 2048×2048, or 4096×4096). Width and height are multiples of 32, and the aspect ratio stays within 1:3 to 3:1.
• Custom WIDTHxHEIGHT: Integer width and height with total pixels from 655,360 to 20,000,000 and an aspect ratio between 1:3 and 3:1 (for example 2560x1440 or 1920x1080). The output image has exactly the requested width and height.

​

Examples

​

Text To Image

{
  "model": "minimax-h3-image",
  "input": {
    "prompt": "A vintage travel poster for Kyoto, headline text 'KYOTO' in large bold serif letters and subtitle 'City of a Thousand Gates' below, red torii gates receding up a mossy hill, autumn maple leaves, 1960s screen-print style",
    "size": "16:9",
    "resolution": "2K"
  }
}

​

Multi-Reference Edit

{
  "model": "minimax-h3-image",
  "input": {
    "prompt": "Place the sofa from <Picture 2> and the floor lamp from <Picture 3> in the empty living room in <Picture 1>. Keep the walls, windows, and daylight of <Picture 1> unchanged.",
    "image_urls": [
      "https://example.com/empty-room.png",
      "https://example.com/sofa.png",
      "https://example.com/floor-lamp.png"
    ],
    "resolution": "2K"
  }
}

​

Custom Size

{
  "model": "minimax-h3-image",
  "input": {
    "prompt": "Minimal presentation background, soft gradient from deep navy to teal, subtle geometric lines in the lower right, large empty area on the left for text",
    "size": "1920x1080",
    "enable_prompt_expansion": false,
    "output_format": "jpeg"
  }
}

​

Notes

• Each request generates one image.
• image_urls must be an array, even for one image.
• Reference images are numbered by their order in image_urls. For best results, give each image a role in the prompt and say what changes and what stays the same.
• A request usually finishes in about 10 to 25 seconds. 4K output and edits with several reference images take longer.

Authorization: Bearer YOUR_API_KEY