Free MiniMax H3 API: video with sound at no cost

What MiniMax H3 Free is

minimax-h3-free is MiniMax H3 run through FastH3 Omni, a distilled version that renders in 8 steps instead of the full schedule. It runs on a GPU we own rather than a paid provider, which is why we can give it away. Every clip has sound generated with the picture: footsteps, voices, music, the room.

It is a separate model id from the paid minimax-h3. Nothing about the paid model changes, and you choose per request which one to call.

How the free queue works

One GPU makes one clip at a time, so free jobs take turns:

• Your job is accepted at once. POST /v1/generations answers 202 with an id, as for any model, and the job shows queued until the GPU has room.
• Turns go by account. When many people are waiting, the GPU takes the next job from the next account in line, so one account sending a burst does not push everyone else back.
• Ten at a time per account. You can have up to 10 free jobs queued or running. An eleventh is refused with 429 concurrency_limit; send it again when one finishes.
• It waits up to three hours. A job still waiting after that fails with model_busy, at no cost.

Once a job reaches the GPU, it renders in about 12 seconds plus 10 seconds per clip second at 768p, so a 5 second clip takes about a minute. Poll every 10 seconds or so; the docs cover every status a job can show.

Generate a free clip from text

The request is the same as for every model on Revana. There is no max_cost_usd to set and no credit to buy first.

export REVANA_API_KEY="sr_live_..."

# 1. Submit the job (returns 202 with the generation id)
curl https://api.revana.si/v1/generations \
  -H "Authorization: Bearer $REVANA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3-free",
    "input": {
      "prompt": "A street drummer plays on upturned buckets on a busy corner at night, coins in a hat at his feet, passers-by slowing to watch. The drums echo off the buildings.",
      "resolution": "768p",
      "aspect_ratio": "9:16",
      "duration": 8
    }
  }'

# 2. Poll with the id from step 1 until status is "succeeded" or "failed"
curl https://api.revana.si/v1/generations/gen_01JC2V8K5T \
  -H "Authorization: Bearer $REVANA_API_KEY"

Start from a picture

Send a starting picture as image_url and the clip opens on it with its people, objects and setting kept. Add up to three more as image_urls (four in all) and name them in the prompt as <Picture 1>, <Picture 2> and so on. Pictures are PNG, JPEG, WebP or GIF at public https links; if you only have a file, upload it first with POST /v1/uploads.

import os
import time

import requests

API = "https://api.revana.si/v1"
HEADERS = {"Authorization": f"Bearer {os.environ['REVANA_API_KEY']}"}

resp = requests.post(f"{API}/generations", headers=HEADERS, json={
    "model": "minimax-h3-free",
    "input": {
        "prompt": "<Picture 1> turns to the camera and smiles, then walks out of frame. Soft cafe chatter.",
        "image_url": "https://example.com/portrait.jpg",
        "duration": 5,
    },
})
resp.raise_for_status()
job = resp.json()

# A free job may wait in the GPU queue before it renders, so poll patiently
while job["status"] in ("queued", "running"):
    time.sleep(10)
    job = requests.get(f"{API}/generations/{job['id']}", headers=HEADERS).json()

if job["status"] == "succeeded":
    print(job["output"]["urls"])
else:
    print("Failed:", job["error"])

Free or paid MiniMax H3?

Use the free model to try ideas, draft storyboards and make social clips where 768p is enough. Use the paid one when you need 2K, a wider frame, first and last frames, more reference pictures or a clip that starts right away.

The full lineup is in the Hailuo and MiniMax guide, and the cheapest video API guide covers every paid model's lowest price. Every model is on the models page, credit packs are on the pricing page, and you can get an API key to run the samples.