MiniMax H3 - AI Model API and Pricing

Built around sound as an input

Most video models generate the image first and leave you to cut it to music. H3 accepts audio references, so a beat drop or a chorus can shape when the shot changes. Reference images and video can go into the same prompt, and one generation can hold several shots. Output is native 2K at 24fps with stereo sound made alongside the picture.

That makes it the natural engine for lyric teasers and artist visualizers. It also suits short social clips where the hook of a track has to land on a visual beat.

Which Ugesi tool to open

AI Music Video Generator is the main home for H3. It takes an image and an audio upload. Presets include Y2K pop, Neon night, Retro film, Anime energy, Dreamy indie and Dark trap. Upload the artist photo and the track, choose a preset, then describe the scene. Try "singer on a rooftop at dusk, wind in her hair, cut to city lights on the chorus".

AI Dance Generator takes one image. Its styles range from Hip-hop groove and K-pop routine to Bollywood and Pet groove. For prompt-only clips, AI Video Generator has presets like Drone reveal and Product turn. Text to Video has Coffee pour and City timelapse. Image to Video offers camera moves such as Orbit and Parallax drift. For the generator, try "slow push in on a perfume bottle on wet black stone, rain starts, soft piano".

Durations, resolution and cost

You can set 5, 6, 8, 10, 12 or 15 seconds. Resolution is 2K only here; MiniMax also runs a cheaper 768p tier, but it is not offered on Ugesi. Since there is no cheaper draft tier, keep test runs short.

A 5-second clip costs 85 credits, about $0.90 at the largest pack. An 8-second clip is 136 credits, about $1.45. The full 15 seconds costs 255 credits, about $2.71. It sits 15th from the cheapest of 31 per-second video models.

When Seedance, Kling or Veo fits better

Seedance 2.5, at 27 credits per second, makes single takes of up to 30 seconds. It adds lip movement for quoted dialogue and accepts up to 50 reference assets. However, it tops out at 720p on this route. Kling 3.0, also 27 credits per second, is stronger on physical action and object handling at 1080p. Its clips come back silent on Ugesi, though. Veo 3.1, at 95 credits per second, is the choice for dialogue-grade masters. Note that the open weights cover only the base model, not the 2K regeneration step.

Where to actually run it

MiniMax H3 API

curl -X POST https://ugesi.com/api/generate \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3",
    "prompt": "a product shot on white marble",
    "seconds": 5
  }'

<?php
$ch = curl_init('https://ugesi.com/api/generate');
curl_setopt_array($ch, [
    CURLOPT_POST => true,
    CURLOPT_RETURNTRANSFER => true,
    CURLOPT_HTTPHEADER => [
        'Authorization: Bearer YOUR_API_KEY',
        'Content-Type: application/json',
    ],
    CURLOPT_POSTFIELDS => json_encode([
        'model' => 'minimax-h3',
        'prompt' => 'a product shot on white marble',
        'seconds' => 5,
    ]),
]);
$response = json_decode(curl_exec($ch), true);
curl_close($ch);
print_r($response);

type GenerateRequest = {
  model: string;
  prompt: string;
  seconds?: number;
};
type GenerateResponse = {
  job_id: number;
  job_token: string;
  credits_held: number;
  status: "queued" | "succeeded";
  output_text?: string;
};

const body: GenerateRequest = {
  model: "minimax-h3",
  prompt: "a product shot on white marble",
  seconds: 5,
};

const resp = await fetch("https://ugesi.com/api/generate", {
  method: "POST",
  headers: {
    "Authorization": `Bearer ${process.env.UGESI_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify(body),
});
if (!resp.ok) throw new Error(`HTTP ${resp.status}`);
const job = (await resp.json()) as GenerateResponse;
console.log(job.job_id, job.status);

import requests

resp = requests.post(
    "https://ugesi.com/api/generate",
    headers={"Authorization": "Bearer YOUR_API_KEY"},
    json={
        "model": "minimax-h3",
        "prompt": "a product shot on white marble",
        "seconds": 5,
    },
    timeout=60,
)
resp.raise_for_status()
print(resp.json())

{
  "model": "minimax-h3",
  "prompt": "a product shot on white marble",
  "seconds": 5
}

The API comes with every credit pack, from Lite to Enterprise, and uses the same balance as the website. API documentation and access →

License and terms of use

Outputs made with paid credits can be used commercially. The maker's own rules for MiniMax H3 also apply (MiniMax API terms). The full rules are in our Terms of service.