Model chooser  /  Generate video  /  Recurring

For video at volume, generate it with MiniMax H3 on fal.

Ad variants and product clips at scale need near-top quality, a simple per-second price and fast generation. H3 Max on fal ranks third on the audio arena, within six Elo points of first, at a lower per-second price than the leader.

How to use it ↓ Try another task

Why MiniMax H3

The reasons it wins for this job.

  • H3 Max scores Elo 1227 on Artificial Analysis's text-to-video with audio arena, six points behind Gemini Omni Flash, and ranks first on image-to-video with audio (1195).
  • fal prices it per second by resolution, with a Turbo mode at half the price, so cost per variant is simple to budget.
  • fal reports a 5-second 768p clip generating in about two and a half seconds, which matters when you render hundreds of variants.
  • The base H3 model has open weights on Hugging Face, so there is a self-hosting path if volume grows, subject to its license.

How to use it

4 steps to a first result.

  1. 1Template the promptFixed camera, style and audio rules; variables for product and setting.
  2. 2Start at 768p and TurboFind prompts that work at the cheap setting, then render winners at 1080p.
  3. 3Review before publishingMotion artefacts are common. A person watches every clip that goes out.
  4. 4Read the license before self-hostingThe open H3 weights exclude some regions and larger companies without written authorization.

fal API, TypeScript

Start from this.

Edit the parts in capitals, then run it.

import { fal } from "@fal-ai/client"; // reads FAL_KEY

const result = await fal.subscribe("minimax/h3-max/text-to-video", {
  input: {
    prompt: "Slow push in on PRODUCT on a desk at golden hour, steam rising from a mug, soft room tone, no text.",
    resolution: "768P",
    duration: 5,
    prompt_expansion_mode: "balanced",
  },
});

console.log(result.data);
Nothing is sent anywhere. It copies to your clipboard.

Alternatives that also work

If MiniMax H3 is not an option.

Wan 3.0Official page →

Alibaba · closed · cost: medium

Pick it when clips must be longer than 15 seconds (up to 30), or you do not need audio.

Gemini Omni FlashOfficial page →

Google · closed · cost: medium

Pick it when you want the top-ranked model and already run on Google Cloud.

Veo 3.1Official page →

Google · closed · cost: high

Pick it when you need Google's enterprise terms and Veo's Fast or Lite tiers for cheaper drafts.

Watch out for

Doing this just once? For a one-off video clip, use Gemini Omni Flash. Gemini Omni Flash ranks first on Artificial Analysis's text-to-video arena for clips with audio, and it is in the Gemini app and Google Flow, so a non-developer can make a clip today.

Sources

Checked . Models change monthly; we re-check this page when they do.

Picking the model is the easy part.

Wiring it into a workflow that runs every week, with evals, fallbacks and a cost you can predict, is the work. Fifteen minutes, no deck.

Book fifteen minutes →