infer
Get started
JUST RELEASED · GENERATIVE VIDEO · MINIMAX H3

MiniMax H3Native 2K. Sound included.

Just released — get unlimited MiniMax H3 for just $49/mo, and try it free for 24 hours. MiniMax's new flagship generates native-2K clips with a soundtrack in the same pass, from text, a keyframe, or up to nine reference images.

OUTPUT · 2K · 5–15sUNLIMITED PLAN · $49 / MO
Try it now
Explosive martial arts action in a courtyard at dusk, dust ring on every impact, whip-pan camera.Generated by MiniMax H3 · MiniMax
Live output
Rain pours through the neon alley, reflections rippling across the wet asphalt, cinematic night grade.
The model turns her head slowly toward camera, studio strobe rim-light, fabric catching the turn.
The spacecraft surges forward through a debris field, engines flaring, hull groaning under strain.
03 / Use cases

Where MiniMax H3 shines.

Cinematic action

Fast, physical sequences — fights, chases, impacts — with the sound of the hit landing in the same generation.

EXAMPLEExplosive martial arts action in a courtyard at dusk, dust ring on every impact, whip-pan camera.

UGC ad creative

Talking-to-camera product clips in vertical, delivery and voice generated together — the format performance teams iterate in volume.

EXAMPLEWoman talking about a cream she loves, handheld selfie framing, bright bathroom light, natural delivery.

Character & creature work

Creatures and characters that hold coherent anatomy through motion — and hold identity across shots by feeding the same reference images (up to nine) into each generation.

EXAMPLEThe dragon slowly raises its head from the treasure hoard, embers drifting, scales catching torchlight.

Fashion & beauty

Editorial motion — skin, fabric, and hair in native 2K where detail is the whole point.

EXAMPLEThe model turns her head slowly toward camera, studio strobe rim-light, fabric catching the turn.

Mood & atmosphere

Noir, rain, neon, smoke — atmosphere-first shots where the generated ambience carries the scene.

EXAMPLEThe lone detective slowly raises his eyes under a flickering streetlamp, rain on the brim of his hat.

Nature & documentary

Wildlife and underwater texture with documentary-style ambient sound, straight out of the model.

EXAMPLESchools of small tropical fish weaving through a sunlit reef, light shafts, ambient underwater hush.

06 / Examples

Generated with MiniMax H3.

A live cross-section of the model's range — portraits, products, typography, illustration, fashion, cinematic. Hover any tile to read its prompt.

Explosive martial arts action in a courtyard at dusk, dust ring on every impact, whip-pan camera.
Rain pours through the neon alley, reflections rippling across the wet asphalt, cinematic night grade.
The spacecraft surges forward through a debris field, engines flaring, hull groaning under strain.
The dragon slowly raises its head from the treasure hoard, embers drifting, scales catching torchlight.
The cat presses a paw into the dough on the counter, kneading in slow rhythm, kitchen ambience.
The lone detective slowly raises his eyes under a flickering streetlamp, rain on the brim of his hat.
Intimate observational portrait: a face in half-light, micro-expressions shifting, shallow depth of field.
The airship cruises above the cloud deck at golden hour, engines humming, pennants snapping in the wind.
The model turns her head slowly toward camera, studio strobe rim-light, fabric catching the turn.
Extreme close-up of a battle-worn warrior's face, sweat and dust in the pores, eyes steadying before the charge.
Schools of small tropical fish weaving through a sunlit reef, light shafts, ambient underwater hush.
Woman talking about a cream she loves, handheld selfie framing, bright bathroom light, natural delivery.
07 / Benchmarks

By the numbers.

Native 2K (2560×1440) — no upscale passOutput resolution
Generated with every clip — no separate sound design stepSoundtrack
Up to 9 images + video + audioReference inputs
VS
Hailuo 02 ProNative 2K with a generated soundtrack, versus 1080p silent clips — the generational jump inside MiniMax's own lineup.
Seedance 2.0 ProSame per-second rate on Infer, at 2K output versus 720p. Seedance keeps the edge on phoneme-level lip-sync.
Veo 3.1 FastHigher resolution (2K vs 1080p) and a longer single-clip ceiling (15s vs 8s). Veo is cheaper per second.
08 / Pricing
$0.13/ second

Pay only for successful generations. No idle, no minimums, no per-seat. Volume discounts kick in at 10K req/mo.

VS NativeSame models, one Infer key: unified SDK across 60+ models, batched billing, no separate MiniMax account.
VS Self-hostClosed weights — self-host isn't an option.
09 / FAQ

Things teams ask.

Q.01What exactly is the $49 offer?
Infer Unlimited — $49/month for unlimited image and video generation in the Studio, MiniMax H3 included, with a 1-day free trial and cancel-anytime. API usage is billed separately, per second of output.
Q.02How long can clips be?
The model generates integer durations from 5 to 15 seconds via the API on Infer (MiniMax's own API documents 4–15s). Studio renders currently run 5-second clips at launch.
Q.03Does it really generate audio?
Yes — every clip ships with a generated soundtrack in the same pass: dialogue, ambience, and effects. All the sample clips on this page carry their original generated audio.
Q.04What resolution is the output?
Native 2K (2560×1440) — it's the model's only output mode. There is no lower-resolution tier to fall back to.
Q.05What input modes are supported?
Three: text-to-video, image-to-video (a first frame, last frame, or both — output aspect follows the image), and reference-to-video, which accepts up to 9 reference images plus up to 3 reference video clips and 3 audio clips in one request.
Q.06Which aspect ratios can it produce?
Text-to-video supports 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. Image-to-video is always adaptive to your input image; reference-to-video defaults to adaptive but accepts any of the fixed ratios.
Q.07How fast is a render?
Measured on Infer: roughly 2.5–4 minutes for a 5-second clip. Generation is queued and async — submit, then poll or use a callback.
Q.08Is there a watermark?
No visible watermark by default. The API exposes an optional AIGC-watermark parameter, off unless you enable it.
Q.09How do I get the best results out of it?
Prompt it like a director: name the camera move (whip-pan, slow push-in, orbit), the light, and the sound you want — the sample prompts on this page all follow that shape. Prompts can run up to 7,000 characters, so there is room to be specific. For a consistent character, attach the same reference images to each generation.
Q.10Can I use the outputs commercially?
Yes. Infer passes through MiniMax's commercial terms, same as Hailuo 02.
10 / Ready to ship?

Ship with MiniMax H3.

One key. One bill. One SDK shape — across 50+ models. Pay only for what you use.

Open in playgroundRead the docs