Start creating with the MiniMax H3 Max AI video studio. Get started →

Independent studio · fal MiniMax H3 Max

MiniMax H3 Max: fast AI video with synced audio

Generate 768p clips at 24 fps with native sound — dialogue, foley, and score in one pass. fal’s post-trained MiniMax H3 Max can denoise a 5-second 768p job in under 3 seconds. This studio exposes every official input and reserves 5 credits per second before the job starts.

768p @ 24 fps 5–15 seconds Native stereo audio T2V · I2V · R2V

Definition

What is MiniMax H3 Max?

MiniMax H3 Max is fal’s fast, post-trained MiniMax H3 variant for 480P and 768P AI video with native stereo audio. H3 Max Studio is an independent workspace at minimaxh3-max.org that calls that model — not MiniMax, Hailuo, or fal. Every second of video costs 5 platform credits.

Read the MiniMax H3 Max guide →

Under 3s GPU denoising for a 5s 768p clip, per fal
Elo 1,201 Artificial Analysis I2V + audio, MiniMax H3 Turbo 768p
768p native 1344×768 at 16:9, plus 480P drafts
5 credits / s Same rate at 480P or 768P on this studio

MiniMax H3 Max AI Video Studio

MiniMax H3 Max · 5 credits per second, reserved before the job starts.

Mode

Refer to files as Image 1, Video 1, Audio 1.

Duration
Resolution
Aspect
Expansion

Image-to-video follows the first frame. No separate aspect control.

At most 12 files total. Extra images after 4 cost 10 credits each. Each reference video costs 45 credits.

Safety checker

Preview / Result

Download

Generated clips

MiniMax H3 Max showcase

Real MiniMax H3 Max videos, streamed from the media CDN — not stills, and not files on this server.

Text to Video · 16:9 · 768P · 15s

1984 countdown

Red type, analog grain, and a rolling leader — picture and a score in one pass.

Text to Video · 16:9 · 768P · 15s

Gregor wakes up

A quiet overhead room, long shadows, and on-screen type that stays locked to the scene.

Text to Video · 16:9 · 768P · 12s

H3 Max spot

A scooter cut against giant type. This is the sample that plays in the studio before you generate.

Text to Video · 16:9 · 768P · 15s

Black-figure vase

A camera move around cracked ceramic, ships, and figures painted on the curve.

Text to Video · 16:9 · 768P · 5s

Tide run

A five-second draft: hydrofoil, spray, and a burning platform on the horizon.

Capabilities

Why MiniMax H3 Max is built for fast iteration

Sub-3s denoising, audio in the same pass, and official first/last-frame plus reference inputs — all wired in this studio.

Speed

Under 3 seconds

Draft a 5-second 768p clip before the coffee cools

fal reports GPU denoising for a 5-second 768p job in under 3 seconds. Queue time still varies. Use short clips to compare camera, light, and sound before you spend a 15-second take.

Try a 5-second draft →

Native audio

Picture and sound together

Dialogue, foley, and beds in the same prompt

MiniMax H3 Max predicts frames and stereo audio together. Write the spoken line, the room, and the score in one field. Preview with sound on before you download.

See prompt formulas →

Director controls

Keyframes and references

First frame, last frame, or up to 12 reference files

Image-to-video follows the still and can land on an end image. Reference-to-video accepts images, video, and audio (audio cannot be the only file). Aspect is official: six ratios on T2V, adaptive plus those six on R2V.

MiniMax H3 Max image-to-video guide →

Independent boards

Where MiniMax H3 Max sits on public leaderboards

Figures below are from public boards, not from this studio. Rankings move. We cite them so you can check the sources.

Artificial Analysis · I2V + audio

Elo 1,201

Listed as MiniMax H3 Turbo (768p). About 2,177 samples, 95% CI ±11. First on that audio-enabled board as of late August 2026.

  • MiniMax H3 Max1,201
  • Nearby frontier models~1,179–1,190

Design Arena · image-to-video

Elo 1,341

Ahead of base MiniMax H3 at 1,333 on the same board. fal’s own stack is what makes Max the fast hosted variant.

  • MiniMax H3 Max1,341
  • MiniMax H3 base1,333

This studio’s price is 5 credits per second, not fal’s launch $0.04/s promo.

Model choice

MiniMax H3 Max vs MiniMax H3

Max is the fast hosted variant this studio calls. Base H3 is the open, slower, 2K-oriented model. We do not run base H3 here. Full H3 Max vs MiniMax H3 comparison.

AreaMiniMax H3 Max (this studio)MiniMax H3 base
GoalFast 768p cloud drafts with native audio2K workflows and local / open weights
Latencyfal: under 3s GPU time for 5s 768pTens of seconds on typical hosted stacks
ResolutionNative 480P and 768P768p base, 2K via a separate upscale path
Inputs on falText, first/last frame, and reference image / video / audioText, multi-file references, editing-style workflows
AudioNative stereo with the pictureNative stereo
What you pay here5 platform credits per secondNot offered on H3 Max Studio

Prompt architecture

How to prompt MiniMax H3 Max

Four parts in one box: subject, camera, light, and sound. Put spoken lines in quotes so the model has something to lip-sync. See the MiniMax H3 Max prompt guide.

1

Subject & action

Who or what moves, and what changes in the shot.

2

Camera

Dolly, track, pan, crane, handheld, or a locked tripod.

3

Light & style

Practicals, neon, overcast, editorial grain, or clean product.

4

Sound

Foley, room tone, music, and any spoken line.

Product spot

Macro shot of a glass perfume bottle on wet black slate. A warm spotlight swings across the label and catches a gold mist. Sound: a single glass tap, then a short aerosol hiss and a low studio bed.

15-second beats

0–5s: wide crane down to a vintage coupe on a dusk beach. 5–10s: the driver checks the mirror. 10–15s: the car pulls away along the wet sand. Sound: surf, a door latch, then a rising engine.

Dialogue close-up

Cinematic close-up of a detective under a fogged street lamp. He says, "The ledger was in the coat the whole time." Sound: slow footsteps on wet stone and one distant bell.

How to use this studio in 3 steps

You stay on minimaxh3-max.org. Credits are reserved up front and returned if fal fails the job.

1

Pick a mode

Text-to-video, image-to-video with first/last frames, or reference-to-video with up to 12 files.

2

Set official fields

Duration 5–15s, 480P or 768P, expansion, seed, safety checker, and the aspect list for that mode. 5 credits per second.

3

Generate and download

Review motion and audio, then download the MP4. Failed jobs refund the reserved credits.

Yearly credit plans

Up to 50% OFF

REFRESHES IN

: :

Studio pricing

Credits for MiniMax H3 Max

This is not fal’s per-second API menu. You buy credits here. Every second of video costs 5 credits — a 5-second clip is 25.

View credit plans

MiniMax H3 Max FAQ

What is MiniMax H3 Max?

MiniMax H3 Max is fal's fast, post-trained MiniMax H3 variant for 480P and 768P AI video with native stereo audio. H3 Max Studio at minimaxh3-max.org is an independent workspace that calls that model. We are not MiniMax, Hailuo, or fal.

How fast is MiniMax H3 Max?

fal reports GPU denoising for a 5-second 768p clip in under 3 seconds. Queue time on this studio still varies. Design Arena and Artificial Analysis also list the hosted Max / H3 Turbo variant first on their image-to-video boards as of late August 2026.

Does MiniMax H3 Max generate audio and dialogue?

Yes. MiniMax H3 Max predicts picture and stereo audio together. Write the spoken line, foley, and music in the same prompt. Preview with sound on before you download.

What is the difference between H3 Max and MiniMax H3?

H3 Max is the fast 480P/768P hosted variant this studio uses. Base MiniMax H3 is the open, slower model aimed at 2K and local setups. We do not run base H3 here.

How much does MiniMax H3 Max cost on this studio?

Every second of video costs 5 platform credits, at 480P or 768P. A 5-second clip is 25 credits. Extra images after the first four cost 10 credits each. Each reference video costs 45 credits.

What resolutions, ratios, and durations does MiniMax H3 Max support?

Official fal inputs: 480P and 768P, 5-15 seconds. Text-to-video ratios are 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. Reference-to-video also has adaptive. Image-to-video follows the first frame.

How do MiniMax H3 Max first and last frames work?

In image-to-video, the first image is the opening frame. An optional last image is the end keyframe. The model interpolates the move between them. Without a first image, fal treats the job as text-to-video.

Can I use MiniMax H3 Max videos commercially?

Yes on paid plans, subject to our Terms and Acceptable Use Policy. You must have rights to every reference you upload. We do not grant MiniMax's or fal's brand rights.

Is MiniMax H3 Max the same as Hailuo?

No. Hailuo is MiniMax's consumer brand. This site is H3 Max Studio, an independent workspace on fal MiniMax H3 Max. We are not Hailuo and we are not MiniMax.

Is there a free MiniMax H3 Max quota here?

Not on this studio. fal's own playground may offer a separate free allowance. Here you buy credits. Failed generations refund the reserved amount.

Generate a 768p clip with audio in this studio

Independent workspace on fal MiniMax H3 Max. Sign in with Google, reserve credits, and keep the MP4 in your history.