MiniMax H3 Image Highspeed API Reference

Generate and edit images with up to 16 references, exact dimensions, 1–10 outputs, and token-based pricing.

Quick example

Submit one image request, save the returned id, and poll until the task is completed or failed. This API uses asynchronous tasks and URL outputs.

curl https://api.seedrouter.ai/v1/images/generations \
  -H "Authorization: Bearer $SEEDROUTER_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"model":"minimax-h3-image-highspeed","prompt":"An amber bottle on pale limestone in soft morning light","area":"1K","aspect_ratio":"3:2"}'

Endpoint and headers

POST https://api.seedrouter.ai/v1/images/generations

Header	Value
Authorization	`Bearer YOUR_API_KEY`
Content-Type	`application/json`

Keep the key on your server. An HTTP success acknowledges a task; it does not mean that its image is ready.

There are 13 accepted fields. Only model and prompt are required. Omit optional fields to use defaults; no field accepts null. The Playground provides the fixed model and ten editable inputs. URL response format and non-streaming task delivery are fixed.

Field	Type	Default	Constraint
`model`	string	Required	`minimax-h3-image-highspeed`
`prompt`	string	Required	1–32000 Unicode characters after trimming
`images`	object array	Omitted	1–16 ordered objects, each containing `image_url`
`disable_prompt_enhanced`	boolean	`false`	Generation only; see dependencies below
`n`	integer	`1`	1–10
`size`	string	`auto`	`auto` or `WIDTHxHEIGHT`; lowercase x
`aspect_ratio`	string	`auto`	`auto` or `X:Y`; X and Y integers 1–300; ratio 1:3–3:1
`area`	string	`4K`	`1K`, `2K`, `4K`; approximate target areas 1024², 2048², 4096² pixels
`output_format`	string	`jpeg`	`jpeg`, `jpg` alias, `png`
`output_compression`	integer	`95`	0–100; JPEG quality; ignored for PNG
`aigc_watermark`	boolean	`false`	Add an AI-generated-content watermark
`response_format`	string	`url`	Only `url`
`stream`	boolean	`false`	Only `false`; poll the task ID

Size and prompt dependencies

Explicit size must contain positive integer dimensions with an area between 655360 and 20000000 pixels, inclusive. Width divided by height must be between 1/3 and 3. For example, 2048x1152 is valid. When using explicit dimensions, omit both aspect_ratio and area, even if their value would be auto.

With omitted or auto size, use aspect_ratio and area together. Area is a target size; inspect the delivered file for exact dimensions. It is not a fixed single-image price.

Setting disable_prompt_enhanced: true requires an explicit pixel size or a fixed aspect ratio. With enhancement disabled, the model reads approximately the first 4096 prompt tokens. The 32000-character submission limit still applies. Editing always enhances prompts, so true is rejected when images is present.

No mask, seed, style, quality, negative prompt, background, or separate resolution field is accepted. Unknown parameters return an error before submission. Streaming, base64 images and WebP output are not offered by this API.

Reference images

Use publicly accessible HTTP(S) URLs without embedded credentials. Each image must be JPEG, PNG, WebP, HEIC or HEIF; at most 64 MiB; 128–8192 pixels on each side; and a long-to-short-side ratio at most 4. Keep the JSON body at or below 64 MiB. Image bytes are sent by URL, not inside the JSON body.

Assign a clear purpose to each reference and describe what should remain unchanged. The array order identifies <Picture 1>, <Picture 2> and so on.

{
  "model": "minimax-h3-image-highspeed",
  "prompt": "Use <Picture 1> as the product and <Picture 2> as the setting. Preserve the product's shape and label. Match the setting's light and perspective.",
  "images": [
    {"image_url": "https://example.com/product.jpg"},
    {"image_url": "https://example.com/setting.jpg"}
  ],
  "aspect_ratio": "3:2",
  "area": "2K",
  "n": 1,
  "output_format": "png"
}

Replace the illustrative URLs with your own accessible image files.

Billing

Billing has three token categories: text input, image input and image output. Multiply actual usage in each category by its live rate. Output count, image dimensions and reference images affect usage; resolution names do not define a fixed bill. Failed requests are not charged. See live pricing.

Task response and polling

An accepted request returns an asynchronous task:

{"id":"task_example","model":"minimax-h3-image-highspeed","status":"processing"}

curl https://api.seedrouter.ai/v1/tasks/task_example \
  -H "Authorization: Bearer $SEEDROUTER_API_KEY"

A completed response contains image URLs in output.data. Each delivered image includes its size and can include revised_prompt (the enhanced description). output.created is the creation time in Unix seconds, and output.output_format identifies the delivered format. output.usage exposes text-input, image-input and image-output token counts and the delivered generated_images count. Example URLs and IDs below are illustrative:

{
  "id": "task_example",
  "model": "minimax-h3-image-highspeed",
  "status": "completed",
  "output": {"data": [{"url": "https://file.deepytb.com/generated/example.jpg"}]}
}

Poll every 10–20 seconds. A client timeout is not a failed task: retain its ID and continue polling instead of submitting a duplicate request. Public responses do not contain account balances or costs.

Errors

HTTP errors use {"error":{"code":integer,"message":string}}. A failed task has status: "failed" and an error. Correct invalid parameters before a new submission. See the shared error directory and task lifecycle.

MiniMax H3 Image Highspeed