MiniMax H3

by MiniMax

Omni-modal video generation by MiniMax (Hailuo). 33B dense DiT that generates video with native synchronized stereo audio (speech, SFX, music) in the same pass. Two checkpoints: FL2VA (t2v / i2v / first-last frame) and Ref2VA (up to 9 reference images + 3 videos + 3 audios lock identity, style and voice). 768p native, 24 fps, up to 15 s. Pruned variants (~20B) cut disk and RAM at equal output.

VideoMiniMax H3 Community LicenseReleased Aug 202633B parameters

Hugging Face repo ↗MiniMax ↗

Drag or click to move along the pipeline

full qualitylightest

Lighter tiers fit smaller GPUs and lose some detail. Start with the heaviest one that loads.

The model

MiniMax H3

BF16from 90.8 GB

Three pieces, three separate downloads.

Diffusion modelGoes in diffusion_models/

Generates the image

Pick one: smaller loads faster, larger keeps more detail.

61.7 GB↓↗

37.5 GB↓↗

Text encoderGoes in text_encoders/

Turns your prompt into guidance

48.0 GB↓↗

VAEGoes in vae/

Decodes the result into pixels

Pick one: smaller loads faster, larger keeps more detail.

4.8 GB↓↗

0.6 GB↓↗

LoRAs (optional)Goes in loras/

1.8 GB↓↗

SamplerKSampler

Copy these values into your KSampler.

Steps8

CFG1

Samplerres_multistep

Schedulersimple

Native res.1344x768 (768p native), 24 fps, up to 15 s

Usage notes

MiniMax H3 Community License. Check the official repo for usage terms.

Click a file name to copy it.