Where can you access MiniMax H3 and start generating videos?

Where can you access MiniMax H3 and start generating?

fal offers the best place to access MiniMax H3 because it's the shortest distance between a prompt box and a finished 2K clip with sound: no hardware to provision, no subscription, and no minimum spend before you see output.

The playground on fal's model page is the one I'd open first, because it puts prompt, duration, resolution, and aspect ratio in a form and shows you the per-second cost before you hit run.

I wrote maybe fifteen prompts there before touching an SDK, which is exactly the right order of operations for a model this sensitive to camera and sound direction.

Then there's the API, where three endpoints cover the range:

• • Text-to-video at minimax/h3/text-to-video takes a prompt and nothing else if you want, defaulting to 5 seconds at 16:9, with 21:9, 4:3, 1:1, 3:4, and 9:16 available and durations running to 15 seconds.
• • Image-to-video at minimax/h3/image-to-video is your first and last frame endpoint, taking a starting image and an optional end image, with aspect ratio inherited from the upload.
• • Reference-to-video at minimax/h3/reference-to-video is where the multimodal pitch stops being marketing, accepting up to 9 images for subject and style, 3 video clips for motion, and 3 audio tracks for voice or score, up to 12 files in one request.

You cite them in the prompt by order, which sounds too-good-to-be-true until you do it once and watch H3 hold a character's face across four shots.

Prompts run to 7,000 characters, so a full shot list with sound notes fits in a single call.

Here’s how an API call looks with fal’s image-to-video endpoint:

import { fal } from "@fal-ai/client";
const result = await fal.subscribe("minimax/h3/image-to-video", {
input: {
prompt: "The camera slowly pulls back from the scene, revealing the full landscape as clouds drift overhead and light shifts across the terrain.",
image_url: "https://storage.googleapis.com/falserverless/example_inputs/hailuo23/pro_i2v_in.jpg"
},
logs: true,
onQueueUpdate: (update) => {
if (update.status === "IN_PROGRESS") {
update.logs.map((log) => log.message).forEach(console.log);
}
},
});
console.log(result.data);
console.log(result.requestId);

What makes fal the best place to use MiniMax H3

What puts fal ahead here is that it was a Day 0 ecosystem partner on the launch, so it had the open weights in hand and built the serving path itself with no third-party access to wait on.

What you're calling is fal's own deployment of H3, not a relayed request to someone else's endpoint.

That's the difference between a host that can tune how a model runs and a middleman who can only pass your request along and add a fee.

It also means the weights stay available to you for research or fine-tuning while the hosted API covers everything you need shipped this quarter.

No GPU provisioning, no queue you have to build, and commercial use is covered on generated output.

The SDK is the same one you'd use for every other model on the platform, so H3 slots into an existing pipeline without a new auth story.

How much does it cost to run MiniMax H3 on fal?

Pricing is per second of output at $0.26, with 2K being the only resolution currently available.

A 5-second clip lands at $1.30, ten seconds at $2.60, and the 15-second maximum at $3.90.

That's pay-per-use with no subscription, no seat licensing, and no minimum commitment, which is the model I'd want for something I'm still exploring.

Run MiniMax H3 on fal

You can open the playground, paste a prompt with actual camera and audio direction in it, and run a 5-second test.

If the shot holds, you can extend the duration and move the same prompt into fal.subscribe.

Once you have reference material worth locking, reference-to-video is where H3 stops feeling like a text-to-video model and starts feeling like a production tool.

That's the sequence that worked for me.