Dialogue-led product spots
Write the line into the prompt and MiniMax H3 Max delivers the mouth, the voice, and the room tone in one render — no separate voice-over pass.
Write one line or drop a single still frame. MiniMax H3 Max hands back a 5–15 second clip with the audio already synced — watch it, change a word, run the next take.
Production preview
Six unedited MiniMax H3 Max takes, text to video and image to video, each with the exact prompt behind it. Copy the shape, change the subject.
"A white kitten chases a butterfly across a sunlit garden. Gentle camera tracking, natural movement, soft afternoon light filtering through the leaves." The sentence carried a subject, a camera move, and a light source — H3 Max staged everything else.
Write a shot like this
A prompt can carry timing, not just subject. This one blocks 0–4s of a warrior-monk crossing a ruined monastery while survey drones sweep the cliffs above, and the beats land where the prompt put them.
Direct a timed shot
Image to video starts from one frame you supply. Hand the portrait a line of dialogue and H3 Max speaks it — the audio arrives synced, not layered on afterwards.
Animate a photo
"Create a pixel-art side-scrolling platformer gameplay animation, presented as authentic 16-bit / 32-bit retro game footage." The prompt asked for a San Francisco endless runner with jump and duck, a SCORE HUD, and chiptune. H3 Max held the sprite, the parallax, and the soundtrack in one pass.
Run a pixel-art take"Stop-motion claymation: a round green frog in a knitted scarf carries a stack of pancakes across a tiny kitchen set, sets it down, then looks at camera and blinks." The prompt called for fingerprints in the clay, felt walls, 12fps stepped motion, and clay squeaks. H3 Max kept the stepped cadence and the sound in the same pass.
Stage a claymation
"Slow-motion close-up of a hummingbird hovering at a red trumpet flower in a sunlit garden, wings a blur, tongue flicking into the bloom." Golden-hour backlight, pollen in the air, one bruised petal. H3 Max held the long-lens bokeh and the rapid wing hum together.
Shoot a slow-motion close-up

MiniMax H3 Max is a video model that turns a written prompt — or a single start frame — into a 5 to 15 second clip that already carries synced audio. It is a tuned build of MiniMax H3, the family MiniMax created for commercial video work: ads, brand films, product spots, UI motion, and game cinematics. In August 2026 evaluations by Artificial Analysis and Design Arena, the tuned build placed first for overall quality, prompt understanding, and aesthetics, and took the top spot on Design Arena's image-to-video board. Choose 480P, 768P or 1080P: a 5-second take costs 35, 75 or 150 credits.
Whole seconds, 5 by default — long enough to carry a beat, short enough to iterate on.
21:9 down to 9:16 for text to video. Image to video simply inherits your start frame.
768P lands at 1344×768, 24 FPS. Iterate at 480P, then re-run the winner at 768P.
Every H3 Max clip renders with sound in place, and MiniMax lists dialogue as stable in 11 languages.
Pin where the shot opens, and optionally where it lands — the motion between is filled in.
MiniMax highlights text and brand rendering, so signs, labels, and logos hold their shape.
One sentence will do — or upload a start frame, plus an end frame if you want to pin the landing.
Choose 5 to 15 seconds, 480P, 768P or 1080P, and one of six aspect ratios — 480P bills 7 credits a second, 768P bills 15 and 1080P bills 30, so a 5-second draft is 35.
The clip comes back with audio in place. Keep it, or change one word and generate again.
Jobs where the finished clip needs to speak, hold brand type, and survive a client review.
Write the line into the prompt and MiniMax H3 Max delivers the mouth, the voice, and the room tone in one render — no separate voice-over pass.
Run the same 15-second spot with dialogue in Japanese, German, or Portuguese; all eleven are listed as stable, so one storyboard covers every market.
Draft the end-card at 480P, then re-render only the winner at 1344×768 — H3 Max keeps lettering readable where earlier models smeared it.
Upload a single headshot as the start frame and get a 5–15 second spoken intro for a landing page or onboarding screen.
Pin the concept art as the first frame and, optionally, the final pose as the last; H3 Max stages the motion and score between them.
Choose 9:16 in text to video and the clip arrives already cropped and voiced for the feed — no reframing in an editor.
Same family, different tuning — figures from Design Arena, August 2026.
| On H3 MaxMiniMax H3 MaxThe tuned build, and what this site runsGenerate | MiniMax H3The base model | |
|---|---|---|
| Tuning | Tuned for prompt adherence and aesthetics | Baseline |
| Design Arena image-to-video Elo (Aug 2026) | 1,341 — ranked #1 | 1,333 |
| Speed at matched quality (Design Arena) | 50×+ the base model | Baseline |
Every MiniMax H3 Max render is billed per second of finished video — 7 credits at 480P, 15 at 768P, 30 at 1080P — and monthly plans top up the balance.
Enjoy Limited-Time 30% OFF!
Signing up grants 20 credits and a 5-second 480P take on MiniMax H3 Max Turbo costs 15, so your first Turbo draft is covered. It stops there: Turbo at 768P is 25 credits and the non-Turbo MiniMax H3 Max at 480P is 35, so the version you keep needs a plan — $19.90 a month loads 1,500 credits.
H3 Max is a tuned build of MiniMax H3 with stronger prompt adherence and better aesthetics. In Design Arena's August 2026 image-to-video rankings it scored Elo 1,341 to the base model's 1,333, taking the top slot.
Each second costs 7 credits at 480P, 15 at 768P or 30 at 1080P. A 5-second take costs 35, 75 or 150 credits; at 15 seconds, 768P costs 225 and 1080P costs 450. Plans start at $19.90 a month for 1,500 credits.
Yes. MiniMax H3 Max generates picture, sound effects, and spoken lines in the same pass, so lip sync is native rather than dubbed on afterwards. MiniMax lists dialogue as stable in 11 languages.
August 2026 benchmarks timed a 5-second 768P MiniMax H3 Max clip at under 3 seconds of model time. At busy hours the queue adds time on top, so treat it as a working pace, not a guarantee.
480P (default), 768P or 1080P. 768P renders at 1344×768 and 24 FPS. MiniMax's hosted base model offers 2K; this generator does not.
The start frame is required and an optional end frame pins where the shot lands. MiniMax H3 Max keeps your still's aspect ratio; the prompt steers motion, camera, and dialogue.
Whole-second lengths from 5 to 15, with 5 as the default, and six frames: 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. 16:9 is the default.
You do, on a paid plan, commercial use included. Clearing third-party rights in your prompt — trademarks, likenesses, copyrighted characters — is on you.
The base model is: MiniMax published H3-Base weights under the MiniMax H3 Community License, not Apache or MIT. The MiniMax H3 Max tuned build this site runs is hosted only, no public weights.
480P to find the shot, 768P for the one you keep — both come back with sound in place.