minimax-h3

Generation is asynchronous: the POST returns a task id, then poll GET /api/v1/tasks/{id} until status is completed.

Quick example

curl https://reapi.ai/api/v1/videos/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3",
    "prompt": "A cat walking slowly on the beach at sunset, cinematic shot, waves gently lapping the shore",
    "aspect_ratio": "16:9",
    "duration": 6
  }'

import requests

resp = requests.post(
    "https://reapi.ai/api/v1/videos/generations",
    headers={
        "Authorization": "Bearer YOUR_API_KEY",
        "Content-Type": "application/json",
    },
    json={
        "model": "minimax-h3",
        "prompt": "A cat walking slowly on the beach at sunset, cinematic shot",
        "aspect_ratio": "16:9",
        "duration": 6,
    },
    timeout=30,
)
print(resp.json())

const r = await fetch("https://reapi.ai/api/v1/videos/generations", {
  method: "POST",
  headers: {
    Authorization: "Bearer YOUR_API_KEY",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "minimax-h3",
    prompt: "A cat walking slowly on the beach at sunset, cinematic shot",
    aspect_ratio: "16:9",
    duration: 6,
  }),
});
console.log(await r.json());

package main

import (
	"bytes"
	"encoding/json"
	"fmt"
	"net/http"
)

func main() {
	body, _ := json.Marshal(map[string]any{
		"model":        "minimax-h3",
		"prompt":       "A cat walking slowly on the beach at sunset, cinematic shot",
		"aspect_ratio": "16:9",
		"duration":     6,
	})
	req, _ := http.NewRequest("POST",
		"https://reapi.ai/api/v1/videos/generations", bytes.NewReader(body))
	req.Header.Set("Authorization", "Bearer YOUR_API_KEY")
	req.Header.Set("Content-Type", "application/json")
	resp, _ := http.DefaultClient.Do(req)
	defer resp.Body.Close()
	var out map[string]any
	json.NewDecoder(resp.Body).Decode(&out)
	fmt.Println(out)
}

Endpoint

POST /api/v1/videos/generations
Authorization: Bearer YOUR_API_KEY
Content-Type: application/json

Submitting returns a task id; poll GET /api/v1/tasks/{id} for the result. Polling does not consume credits.

Parameters

* In I2V, provide first_frame_url, last_frame_url, or both. ** In R2V, provide reference_image_urls or reference_video_urls (audio alone is not accepted).

Frames (first_frame_url / last_frame_url) and reference inputs (reference_*_urls) are mutually exclusive. All media inputs must be public HTTP(S) URLs — base64 / data: URIs are rejected.

Modes

Mode is implicit — selected by which inputs you send:

// I2V — animate between two frames
{ "model": "minimax-h3", "prompt": "the character turns and smiles, camera pushes in", "first_frame_url": "https://…/first.jpg", "last_frame_url": "https://…/last.jpg", "duration": 6 }

// R2V — identity from images, motion from a clip, voice from audio
{ "model": "minimax-h3", "prompt": "a continuous cinematic scene using the referenced character, motion, and voice", "reference_image_urls": ["https://…/id.jpg"], "reference_video_urls": ["https://…/motion.mp4"], "reference_audio_urls": ["https://…/voice.mp3"], "duration": 8 }

Pricing

Billed per second at one flat 2K rate; the routing mode does not change the rate. Three components:

• Output seconds — the requested duration.
• Input video seconds — the server-probed total duration of reference_video_urls is added to the billable seconds (the vendor processes input and output footage alike).
• Extra reference images — the first 5 are free; images 6–9 bill a flat per-image rate. Audio input is free.

See the model page for current rates.

Bill formula (1 credit = $0.001):

credits = ceil((per_second_usd × (duration + input_video_seconds)
        + extra_image_usd × max(0, reference_images − 5)) × 1000)

Output

On success, GET /api/v1/tasks/{id} returns:

{
  "id": "task_…",
  "model": "minimax-h3",
  "status": "completed",
  "output": { "video_urls": ["https://cdn.reapi.ai/media/tasks/…/0.mp4"] },
  "error": null
}

The clip includes its native stereo audio track in the MP4.

Errors

Failed generations are surfaced under error in the polling response and are refunded automatically. Full catalog: Errors.

Tips

• Direct the audio in the prompt. MiniMax H3 renders sound with the picture — describe dialogue lines, music mood, ambience, and effect cues alongside the visual action for the tightest sync.
• Assign roles to references. In R2V, say what each asset controls (identity, motion, camera language, voice); it reduces conflicts between reference materials.
• Watch reference-video length. Input clips bill their own seconds on top of the output — a 12s reference plus a 10s output bills as 22s.
• aspect_ratio is rejected in I2V — crop your frames to the orientation you want instead.
• Longer duration scales the bill linearly; start at the 6s default while iterating.

Related

• Tasks — universal polling endpoint
• Errors — full error catalog
• seedance-2-0 — sibling multimodal video family
• veo3-1 — premium video generation with optional audio