MiniMax H3

POST

tasks

MiniMax H3

curl --request POST \
  --url https://api.qingbo.ai/v1/tasks \
  --header 'Content-Type: application/json' \
  --data '
{
  "model": "<string>",
  "action": "<string>",
  "prompt": "<string>",
  "duration": 123,
  "resolution": "<string>",
  "aspect_ratio": "<string>",
  "image_urls": [
    "<string>"
  ],
  "first_frame_image": "<string>",
  "last_frame_image": "<string>",
  "video_urls": [
    "<string>"
  ],
  "audio_urls": [
    "<string>"
  ],
  "image_with_roles": [
    {}
  ],
  "watermark": true,
  "ref_task_id": "<string>"
}
'

import requests

url = "https://api.qingbo.ai/v1/tasks"

payload = {
    "model": "<string>",
    "action": "<string>",
    "prompt": "<string>",
    "duration": 123,
    "resolution": "<string>",
    "aspect_ratio": "<string>",
    "image_urls": ["<string>"],
    "first_frame_image": "<string>",
    "last_frame_image": "<string>",
    "video_urls": ["<string>"],
    "audio_urls": ["<string>"],
    "image_with_roles": [{}],
    "watermark": True,
    "ref_task_id": "<string>"
}
headers = {"Content-Type": "application/json"}

response = requests.post(url, json=payload, headers=headers)

print(response.text)

const options = {
  method: 'POST',
  headers: {'Content-Type': 'application/json'},
  body: JSON.stringify({
    model: '<string>',
    action: '<string>',
    prompt: '<string>',
    duration: 123,
    resolution: '<string>',
    aspect_ratio: '<string>',
    image_urls: ['<string>'],
    first_frame_image: '<string>',
    last_frame_image: '<string>',
    video_urls: ['<string>'],
    audio_urls: ['<string>'],
    image_with_roles: [{}],
    watermark: true,
    ref_task_id: '<string>'
  })
};

fetch('https://api.qingbo.ai/v1/tasks', options)
  .then(res => res.json())
  .then(res => console.log(res))
  .catch(err => console.error(err));

package main

import (
	"fmt"
	"strings"
	"net/http"
	"io"
)

func main() {

	url := "https://api.qingbo.ai/v1/tasks"

	payload := strings.NewReader("{\n  \"model\": \"<string>\",\n  \"action\": \"<string>\",\n  \"prompt\": \"<string>\",\n  \"duration\": 123,\n  \"resolution\": \"<string>\",\n  \"aspect_ratio\": \"<string>\",\n  \"image_urls\": [\n    \"<string>\"\n  ],\n  \"first_frame_image\": \"<string>\",\n  \"last_frame_image\": \"<string>\",\n  \"video_urls\": [\n    \"<string>\"\n  ],\n  \"audio_urls\": [\n    \"<string>\"\n  ],\n  \"image_with_roles\": [\n    {}\n  ],\n  \"watermark\": true,\n  \"ref_task_id\": \"<string>\"\n}")

	req, _ := http.NewRequest("POST", url, payload)

	req.Header.Add("Content-Type", "application/json")

	res, _ := http.DefaultClient.Do(req)

	defer res.Body.Close()
	body, _ := io.ReadAll(res.Body)

	fmt.Println(string(body))

}

HttpResponse<String> response = Unirest.post("https://api.qingbo.ai/v1/tasks")
  .header("Content-Type", "application/json")
  .body("{\n  \"model\": \"<string>\",\n  \"action\": \"<string>\",\n  \"prompt\": \"<string>\",\n  \"duration\": 123,\n  \"resolution\": \"<string>\",\n  \"aspect_ratio\": \"<string>\",\n  \"image_urls\": [\n    \"<string>\"\n  ],\n  \"first_frame_image\": \"<string>\",\n  \"last_frame_image\": \"<string>\",\n  \"video_urls\": [\n    \"<string>\"\n  ],\n  \"audio_urls\": [\n    \"<string>\"\n  ],\n  \"image_with_roles\": [\n    {}\n  ],\n  \"watermark\": true,\n  \"ref_task_id\": \"<string>\"\n}")
  .asString();

MiniMax H3 is MiniMax’s omni-modal video model: one endpoint takes text, image, video and audio input. This page covers three models — minimax-h3 renders 768P / 2K clips with their own soundtrack; minimax-h3-max is the high-quality edition at 480P / 768P / 1080P; minimax-h3-regeneration turns a completed minimax-h3 768P clip into native 2K. For plain text-to-video or first-frame video without multimodal references, the Hailuo Series costs less.

Quick start

curl -X POST https://api.qingbo.ai/v1/tasks \
  -H "Authorization: Bearer $WAVE_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: h3-demo-001" \
  -d '{
    "model": "minimax-h3",
    "action": "generate",
    "prompt": "A mountain town at dawn, mist lifting as the camera pushes from the rooftops toward the peaks",
    "duration": 8,
    "resolution": "2k",
    "aspect_ratio": "16:9"
  }'

curl -X POST https://api.qingbo.ai/v1/tasks \
  -H "Authorization: Bearer $WAVE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3",
    "action": "generate",
    "prompt": "Day turns to night as the lights come on one by one",
    "first_frame_image": "https://cdn.example.com/day.jpg",
    "last_frame_image": "https://cdn.example.com/night.jpg",
    "duration": 10,
    "resolution": "768p"
  }'

curl -X POST https://api.qingbo.ai/v1/tasks \
  -H "Authorization: Bearer $WAVE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3",
    "action": "generate",
    "prompt": "Keep the camera rhythm of the reference video and move the scene to snowy mountains",
    "image_urls": ["https://cdn.example.com/character.jpg"],
    "video_urls": ["https://cdn.example.com/reference.mp4"],
    "audio_urls": ["https://cdn.example.com/bgm.mp3"],
    "duration": 12,
    "resolution": "2k"
  }'

curl -X POST https://api.qingbo.ai/v1/tasks \
  -H "Authorization: Bearer $WAVE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3-max",
    "action": "generate",
    "prompt": "The character from the reference images walks into the street from the reference video, tracking shot",
    "image_urls": [
      "https://cdn.example.com/character.jpg",
      "https://cdn.example.com/outfit.jpg"
    ],
    "video_urls": ["https://cdn.example.com/street.mp4"],
    "duration": 8,
    "resolution": "1080p",
    "aspect_ratio": "16:9"
  }'

curl -X POST https://api.qingbo.ai/v1/tasks \
  -H "Authorization: Bearer $WAVE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3-regeneration",
    "action": "regenerate",
    "ref_task_id": "task-wave1788783825b950188816"
  }'

A successful submission returns a task_id. Retrieve the result with GET /v1/tasks/{task_id}, wait inline with Prefer: wait, or configure a webhook. Chaining a regeneration: generate a clip with minimax-h3 at resolution: "768p"; once the task is completed, pass the returned task_id as ref_task_id to minimax-h3-regeneration to get the same content in native 2K. ref_task_id is a task ID from this platform, not an upstream ID; the prompt and reference material come from the source task and are not resubmitted.

Request parameters

string

required

minimax-h3, minimax-h3-max or minimax-h3-regeneration.

string

default:"generate"

generate: the only action of minimax-h3 and minimax-h3-max
regenerate: the only action of minimax-h3-regeneration, requires ref_task_id

string

required

Description of the shot; required for generate, including when reference material is attached.

integer

required

Clip length in seconds, an integer; required for generate. 4–15 on minimax-h3, 5–15 on minimax-h3-max.

string

Output resolution:

minimax-h3: 768p or 2k, default 2k; regeneration with minimax-h3-regeneration requires a 768p source task
minimax-h3-max: 480p, 768p or 1080p, default 768p

string

Frame ratio: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16. With a first or last frame the ratio follows the image.

string[]

Reference image URLs, up to 9. Every image is treated as a reference image; they are never mapped to first / last frames by count.

string

URL of the first frame, used as the opening image of the clip. Cannot be mixed with image_urls, video_urls or audio_urls.

string

URL of the last frame; combine with first_frame_image for an A-to-B transition. Cannot be mixed with image_urls, video_urls or audio_urls.

string[]

Reference video URLs for camera work, rhythm and style. Up to 3 clips, 15 seconds in total, 23.976–60 fps; on minimax-h3-max each clip is 2–15 seconds.

string[]

Reference audio URLs, up to 3 clips, 15 seconds in total.

minimax-h3: must be sent together with a reference image
minimax-h3-max: must be sent together with a reference image or a reference video; each clip is 2–15 seconds

array

Role-tagged images of {url, role}, role being first_frame, last_frame or reference_image. At most 2 items and not together with image_urls; send further reference images through image_urls. minimax-h3-max only.

boolean

default:"false"

Add a watermark to the output. Cannot be sent at 1080p on minimax-h3-max.

string

ID of a completed minimax-h3 task. Used — and required — only by the regenerate action of minimax-h3-regeneration. The source must be a successful minimax-h3 generate task on the same account at 768p.

The shared callback_url, callback_events, Prefer: wait, Idempotency-Key and cost-cap headers are documented in Submit Task. minimax-h3 and minimax-h3-max do not accept seed, n, negative_prompt, quality or generate_audio. minimax-h3-regeneration accepts only model and ref_task_id; everything else is inherited from the source task. Anything beyond the parameters above returns 400 and is not billed.

Limits

Model	Condition	Limit
`minimax-h3` `minimax-h3-max`	`action: "generate"`	A prompt is required.
`minimax-h3`	`action: "generate"`	Video generation requires an explicit duration.
`minimax-h3`	`first_frame_image` sent	Frame mode cannot be mixed with multimodal references.
`minimax-h3`	`audio_urls` sent	Reference audio must be provided together with a reference image.
`minimax-h3-max`	`first_frame_image` or `last_frame_image` sent	Frame mode cannot be mixed with multimodal references.
`minimax-h3-max`	`audio_urls` sent without reference images	Reference audio must be sent together with a reference image or a reference video.
`minimax-h3-max`	Every request	At most 3 reference videos and 3 reference audio clips; image_with_roles takes at most 2 items.
`minimax-h3-max`	`image_with_roles` sent	image_with_roles cannot be combined with image_urls; send reference images through image_urls.
`minimax-h3-max`	`resolution: "1080p"`	1080p does not take the watermark parameter; omit it.
`minimax-h3-regeneration`	`action: "regenerate"`	Regeneration requires a completed local H3 task id whose output is 768p (pass resolution=768p on generate).

Item	`minimax-h3`	`minimax-h3-max`
Duration	4–15 seconds	5–15 seconds
Reference images	Up to 9	Up to 9
Reference video	Up to 3 clips, 15 s in total	Up to 3 clips, 2–15 s each, 15 s in total
Reference audio	Up to 3 clips, 15 s in total	Up to 3 clips, 2–15 s each, 15 s in total

Pricing

minimax-h3 and minimax-h3-max are billed by resolution tier times seconds: each resolution tier has a per-second rate, multiplied by duration. Two more dimensions apply only when used:

Reference video: with video_urls, output seconds are charged at the reference rate of the same resolution, and the actual total length of the reference video is charged on top at the input-second rate. When the length cannot be read, 15 seconds is used.
Reference images: the first images in image_urls add nothing (5 on minimax-h3, 2 on minimax-h3-max); each further image adds a fixed amount. First frames, last frames and reference audio are not charged separately.

minimax-h3-regeneration follows the same rules at the 2k rate, with the billed duration and reference material taken from the source task behind ref_task_id; reference video and reference images of the source task are charged again on regeneration.

Per-model rates are in price_config on GET /v1/models, and in the console’s Model Market.What a single call actually cost is the cost field on the task response — an integer quota at 500,000 quota = 1 USD.

You are only billed for a video that is produced. Failed and cancelled tasks, and tasks that return no usable video, are refunded in full. The final charge is the cost field on the task response — an integer quota, not dollars.

Response

Completed task

{
  "task_id": "task-wave1788783825b950188816",
  "model": "minimax-h3",
  "action": "generate",
  "status": "completed",
  "progress": "100%",
  "created_at": 1788783825,
  "completed_at": 1788783934,
  "result": {
    "videos": [
      {
        "url": ["https://cdn.example.com/result.mp4"],
        "expires_at": 1789089508
      }
    ]
  },
  "billing_status": "settled",
  "cost": 144000,
  "urls": {
    "get": "https://api.qingbo.ai/v1/tasks/task-wave1788783825b950188816",
    "cancel": "https://api.qingbo.ai/v1/tasks/task-wave1788783825b950188816/cancel"
  }
}

result.videos holds the finished clip; expires_at is when the link expires (Unix seconds), so copy the file to your own storage before then. minimax-h3-max and regeneration tasks return the same shape. See Query Task Status for every field.

Available models

Model ID	Action	Resolutions	Duration (s)	Input
`minimax-h3`	`generate`	`768p` `2k`	4–15	Text, reference images, first / last frames, reference video, reference audio
`minimax-h3-max`	`generate`	`480p` `768p` `1080p`	5–15	Text, reference images, first / last frames, reference video, reference audio
`minimax-h3-regeneration`	`regenerate`	`2k`	From the source task	ID of a completed `minimax-h3` 768P task

MiniMax H3

curl --request POST \
  --url https://api.qingbo.ai/v1/tasks \
  --header 'Content-Type: application/json' \
  --data '
{
  "model": "<string>",
  "action": "<string>",
  "prompt": "<string>",
  "duration": 123,
  "resolution": "<string>",
  "aspect_ratio": "<string>",
  "image_urls": [
    "<string>"
  ],
  "first_frame_image": "<string>",
  "last_frame_image": "<string>",
  "video_urls": [
    "<string>"
  ],
  "audio_urls": [
    "<string>"
  ],
  "image_with_roles": [
    {}
  ],
  "watermark": true,
  "ref_task_id": "<string>"
}
'

import requests

url = "https://api.qingbo.ai/v1/tasks"

payload = {
    "model": "<string>",
    "action": "<string>",
    "prompt": "<string>",
    "duration": 123,
    "resolution": "<string>",
    "aspect_ratio": "<string>",
    "image_urls": ["<string>"],
    "first_frame_image": "<string>",
    "last_frame_image": "<string>",
    "video_urls": ["<string>"],
    "audio_urls": ["<string>"],
    "image_with_roles": [{}],
    "watermark": True,
    "ref_task_id": "<string>"
}
headers = {"Content-Type": "application/json"}

response = requests.post(url, json=payload, headers=headers)

print(response.text)

const options = {
  method: 'POST',
  headers: {'Content-Type': 'application/json'},
  body: JSON.stringify({
    model: '<string>',
    action: '<string>',
    prompt: '<string>',
    duration: 123,
    resolution: '<string>',
    aspect_ratio: '<string>',
    image_urls: ['<string>'],
    first_frame_image: '<string>',
    last_frame_image: '<string>',
    video_urls: ['<string>'],
    audio_urls: ['<string>'],
    image_with_roles: [{}],
    watermark: true,
    ref_task_id: '<string>'
  })
};

fetch('https://api.qingbo.ai/v1/tasks', options)
  .then(res => res.json())
  .then(res => console.log(res))
  .catch(err => console.error(err));

package main

import (
	"fmt"
	"strings"
	"net/http"
	"io"
)

func main() {

	url := "https://api.qingbo.ai/v1/tasks"

	payload := strings.NewReader("{\n  \"model\": \"<string>\",\n  \"action\": \"<string>\",\n  \"prompt\": \"<string>\",\n  \"duration\": 123,\n  \"resolution\": \"<string>\",\n  \"aspect_ratio\": \"<string>\",\n  \"image_urls\": [\n    \"<string>\"\n  ],\n  \"first_frame_image\": \"<string>\",\n  \"last_frame_image\": \"<string>\",\n  \"video_urls\": [\n    \"<string>\"\n  ],\n  \"audio_urls\": [\n    \"<string>\"\n  ],\n  \"image_with_roles\": [\n    {}\n  ],\n  \"watermark\": true,\n  \"ref_task_id\": \"<string>\"\n}")

	req, _ := http.NewRequest("POST", url, payload)

	req.Header.Add("Content-Type", "application/json")

	res, _ := http.DefaultClient.Do(req)

	defer res.Body.Close()
	body, _ := io.ReadAll(res.Body)

	fmt.Println(string(body))

}

HttpResponse<String> response = Unirest.post("https://api.qingbo.ai/v1/tasks")
  .header("Content-Type", "application/json")
  .body("{\n  \"model\": \"<string>\",\n  \"action\": \"<string>\",\n  \"prompt\": \"<string>\",\n  \"duration\": 123,\n  \"resolution\": \"<string>\",\n  \"aspect_ratio\": \"<string>\",\n  \"image_urls\": [\n    \"<string>\"\n  ],\n  \"first_frame_image\": \"<string>\",\n  \"last_frame_image\": \"<string>\",\n  \"video_urls\": [\n    \"<string>\"\n  ],\n  \"audio_urls\": [\n    \"<string>\"\n  ],\n  \"image_with_roles\": [\n    {}\n  ],\n  \"watermark\": true,\n  \"ref_task_id\": \"<string>\"\n}")
  .asString();

Text Generation

Image Generation

Video Generation

Audio

Tools & Embeddings

Task Management

Account & Usage

Quick start

Request parameters

Limits

Pricing

Response

Available models

​Quick start

​Request parameters

​Limits

​Pricing

​Response

​Available models

​Related

Quick start

Request parameters

Limits

Pricing

Response

Available models

Related