VIDEO + SOUND · MINIMAX-H3

Ideas in motion.

Explore the DMAD project’s example videos, or try a short text-to-video generation via a hosted inference provider.

4-step adapterNative audioHosted inference

Create a clip Experimental

Uses WaveSpeed through Hugging Face. The Space does not host model weights. A provider-enabled token and available credits are required.

Describe your scene

Resolution

Aspect ratio

Hugging Face inference token

Use an Inference Providers token ↗. The token stays in this page’s memory and is sent to Hugging Face's router; the prompt goes to WaveSpeed.

A single request asks for 5 seconds. Provider charges may apply to your account. Check current pricing ↗ and account limits before submitting. Failures or retries may be billed.

I approve one potentially paid request and sending my prompt to the provider.

No request submitted.

The DMAD adapter is specified, but this provider integration and its sampler have not been verified with a live generation. Output and language accuracy are not guaranteed.

Explore the examples

Original samples from the DMAD authors—not clips generated by this Space.

AUTHOR'S SAMPLE

Open sample directly ↗

4 / 8denoising steps

5–15sexample clips

Audiojoint generation

ComfyUI workflows ↗Project code ↗Read the paper ↗

The reference workflow's 8 GB GPU mode still recommends 64 GB system RAM. Your 2 GB GPU does not meet that demonstrated setup.