MiniMax H3
by MiniMax
Reference-driven video that keeps your song in the picture.
Native stereo sound, 768P up to 4K.
Key features
Your song is the soundtrack
Reference audio you supply comes back as the output track, so the take lands already on your master, not on invented sound.
Reference to video
Hold characters and scenes across shots with up to nine reference images, plus clips and audio stems, so a three-shot chorus stays one world.
Four to fifteen seconds
Each take runs four to fifteen seconds, six by default — the length a chorus cut actually needs.
Seven ratios
21:9 through 9:16 plus adaptive, so the same brief covers cinema wides and verticals for release day.
Up to 4K
768P is the working resolution, 2K unlocks on Pro, and Creator renders at up to 4K for the final master.
Native stereo
Picture and sound are predicted together, so mouth movement and beats line up with the audio you handed the model.
Whole songs made with MiniMax H3
Finished music videos, each cut to a full song and its original audio. Open one to see its shots, the models that made them and the credits it took.
Technical specifications
MiniMax H3
Reference-driven video model, served through dimly
Reference to video
Images, clips and audio drive the take
4 to 15s
Six seconds by default, the chorus-cut length
768P to 4K
Every resolution on every plan
7 ratios
21:9, 16:9, 4:3, 1:1, 3:4, 9:16, adaptive
Native stereo
Your reference audio returns as the output track
9 references
Plus up to 3 clips and 3 audio stems per take
Beat-locked
Motion follows the audio you supply, take after take
Prompt examples
Neon chorus
A singer turns to the camera as the chorus hits, neon signs flickering behind her
Edit promptVinyl close-up
Extreme close-up of a needle dropping onto spinning vinyl, dust in the light
Edit promptUse cases

Chorus-cut three-shots
The three-shot loop around your chorus, directed as one world instead of three unrelated clips.

Previz before the shoot
Block the video, watch it, adjust it, watch it again — with takes cheap enough to explore a doubtful idea.

Instrumentals and post-rock
No singer needed — light, weather and a slow camera carry a track on their own.

Character consistency
Keep the same artist across every shot of a sequence by feeding the same reference set.

Sound-led scenes
Scenes built around a precise moment — a snare, a door, a first word — arrive already carrying it.

Era and film stock
Name the decade, the lens or the stock — a 16mm gym show, a VHS camcorder dusk — and the take arrives looking like it.
Overview
MiniMax H3 is the reference-driven video model behind dimly's music-video pipeline. You hand it a brief plus references — up to nine images, three clips and three audio stems — and it returns a four-to-fifteen-second take with stereo sound generated alongside the picture. When you supply reference audio, that audio comes back as the output track, which is why a dimly sequence lands already on your song.
What the reference input buys you
References are what keep a music video feeling like one film. The same character images and the same world stills feed every take in a sequence, so shot one and shot three read as the same artist in the same room. Audio references do the timing work: motion follows the track you supplied, so a snare hit or a sung line lands where the song puts it.
The trade is deliberate: H3 is a reference-first model, not a raw text-to-video toy. Write the brief across the clip — subject, action, camera, light, sound — and let the references carry identity.
Where Seedance 2.5 is the better pick
Seedance 2.5 takes first and last frames, so it is the tool when you know exactly where a move starts and ends. It runs four to thirty seconds against H3's fifteen, at 480p and 720p, and dimly runs it with its own audio generation switched off — the original song is muxed back after the render, so the master is never touched.
In practice the pairing is simple: direct character-led chorus work on H3, and use Seedance when a move needs exact endpoints or a longer runway.
MiniMax H3 on dimly
H3 sits in dimly's model picker on every plan, at 768P, 2K or 4K — higher resolutions simply cost more credits per second. Every generation is quoted in credits and confirmed by you before it runs, and takes land in the project workspace where you can revise, re-time, or send the same brief through Seedance to compare.
Simple pricing
Plan your video for free, then pick a plan that fits your release schedule.
Starter
A steady month of chorus cuts, or one shorter song.
- 600 credits, valid for 30 days
- ≈ 8 chorus cuts (15 s)
- or one song up to 2:20
- Every model and resolution · 3 renders at once
Pro
One full song a month with room to retake, and bigger sizes when a release needs them.
Credits a month
- 1,000 credits, valid for 30 days
- ≈ 14 chorus cuts (15 s)
- or 1 full song (3:45)
- Every model and resolution · 4 renders at once
Creator
Several full songs a month, for a steady release schedule.
- 5,000 credits, valid for 30 days
- ≈ 72 chorus cuts (15 s)
- or ≈ 5 full songs (3:45)
- Every model and resolution · 8 renders at once
Full song, chorus cut, lyric video or visualizer · Watermark-free exports on every plan · Export rights depend on your source permissions and provider terms
Credit packs
Pay once, no subscription. Made for one AI effect or one song; if you will make more, a plan above is the better rate.
Single
One AI effect video, like Hotel Lobby or AI Rap Duo.
- 250 credits, valid for 1 year
- 1 AI effect video (Hotel Lobby, AI Rap Duo)
- or ≈ 3 chorus cuts (15 s)
- Exports carry the Dimly watermark
- No subscription, never renews
Song
One full music video (a song up to 3:45), or four AI effects.
- 1,000 credits, valid for 1 year
- 4 AI effect videos (Hotel Lobby, AI Rap Duo)
- or 1 full song (3:45)
- Watermark-free exports
- No subscription, never renews
Studio
A full song in 2K with room for retakes, or two songs.
- 2,000 credits, valid for 1 year
- 8 AI effect videos (Hotel Lobby, AI Rap Duo)
- or ≈ 2 full songs (3:45)
- Watermark-free exports
- No subscription, never renews
Estimates use MiniMax H3 at 768P with 5-second shots (a 15-second chorus cut counts as three), a keyframe per shot, two reference sheets and no retakes; a full song means 3:45, the typical length of the songs people bring. Higher resolutions and the other models cost more per video. You see the exact quote before anything renders, and failed renders are refunded.
FAQs
It is dimly's reference-driven video model: you supply a brief plus up to nine reference images, three clips and three audio stems, and it returns a four-to-fifteen-second take with native stereo sound. It powers dimly's music videos, from a chorus cut to the whole song, on every plan.
Yes. When you attach reference audio, H3 returns that audio as the output track, and motion follows it — the take lands already on your master rather than on sound the model invented.
Every paid plan and credit pack, at every resolution: 768P, 2K or 4K. Higher resolutions cost more credits per second; generation is quoted in credits and confirmed before it runs.
Four to fifteen seconds per take, six by default. A chorus cut is built from several short takes rather than one long render, which is what keeps revision loops fast.
H3 is reference-first: identity and timing come from the images, clips and audio you attach, and the audio you supply returns as the track. Seedance takes first and last frames, runs up to thirty seconds at 480p/720p, and renders silent in dimly so your original song is muxed back afterwards.
Seven: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 and adaptive, so the same references feed cinema wides, square posts and release-day verticals.
Every render is quoted before it runs — you see the credit cost and confirm it. Credits expire with your billing cycle, and planning the video itself is free.
Other models
See all modelsSeedance 2.5
ByteDance
First-and-last-frame moves up to thirty seconds, audio-driven at 480p/720p — your song muxed back intact.
Kling 4.0
Kuaishou
Kuaishou's long-take video model: up to thirty seconds in one pass, ten keyframes and Omni Reference with up to fifteen inputs.
MiniMax H3 Image
MiniMax
MiniMax's open-weights image model: up to sixteen reference images, native 4K and Chinese or English lettering, in about twenty seconds.
Seedream 5.5
ByteDance Seed
ByteDance Seed's image model: natural skin and light, edits that leave the rest of the frame alone, and dense small text that stays legible.












