What MiniMax H3 is
MiniMax H3 (third generation of the Hailuo line) is a video model by China's MiniMax, released in August 2026. What sets it apart from earlier versions: sound is generated together with the picture. The model synthesises a stereo track in the same pass as the video, so footsteps, breathing, city noise and lines land in sync — no separate scoring or sound design pass.
Clips run 4 to 15 seconds at 768p or 2K, with seven aspect ratios including vertical 9:16 for Reels and Shorts.
On Neuromia, MiniMax H3 runs online in the browser — nothing to install.
Three ways to drive a generation
From text. Describe the scene, the action and the mood, and the model composes the shot. The aspect ratio is chosen here too: 16:9 for YouTube, 9:16 for vertical feeds, 21:9 for a cinematic frame.
From frames. Upload a first frame and H3 brings the still to life. Add a last frame and the model builds the transition between the two images. Proportions come from the picture, so nothing gets cropped.
From references. The most interesting mode: up to 9 photos, 3 videos and 3 audio files in a single request. Photos hold a character's appearance or a style across the clip, video sets the kind of motion and camera work, audio is a sample of how the result should sound. References never become frames and are not cropped — the model treats them as samples.
How to use MiniMax H3
- Open Neuromia in your browser — nothing to download.
- Sign up by e-mail and get starter credits.
- Pick MiniMax H3 in the video section.
- Choose the scenario — Text, Frames or References — then set duration, aspect ratio and quality.
- Wait for the render (usually 2–5 minutes) and download the clip with its soundtrack.
Pricing
You pay in Neuromia credits, and the price is shown before you start — you cannot go negative. Billing is per second and depends on quality: 768p is the cheaper tier, 2K is roughly 1.6× more. A 4-second clip at 768p starts at 99 credits.
H3 versus Hailuo 2.3
Hailuo 2.3 is the previous MiniMax generation, strong at photo animation: expressive faces, natural portrait motion. If that is your task, see the Hailuo page.
H3 solves a wider problem: a longer clip with sound, multimodal input and character consistency driven by references. Hailuo still owns portrait animation; H3 owns the scene with a soundtrack.
Things to keep in mind
Clips cap at 15 seconds, so a longer story has to be assembled from several generations. Audio is always on — the model has no toggle for it, and an uploaded audio file is not laid over the result: it only guides how the generated sound behaves. Budget more credits and render time for 2K.
Bottom line
MiniMax H3 on Neuromia means video with stereo sound up to 2K and 15 seconds, three ways to define the scene, and a browser-based workflow. Your first clip runs on starter credits.
Scenario walkthrough
A step-by-step guide to animating family and vintage photographs — «bring an old family photo to life with AI».