MiniMax AI video in 2K with sound
MiniMax H3 AI Video Generator
MiniMax H3 puts clean, correctly spelled words on screen, from a perfume title card to a word built out of pouring sand, in 2K video with sound in the same file.
MiniMax H3 is the next-generation video model from MiniMax, the team behind Hailuo. On Pixel Dojo it turns a text prompt, a first frame or reference images into 5 to 15 second clips at 2K or 768p, with sound built in. We made every MiniMax H3 clip on this page on Pixel Dojo, and each one comes with its exact prompt so you can start from it.
20 credits ($1.25) per 5 s clip at 2K. Included in every plan, from $10/month.
Already a member? Sign in
By Brian Lovett · Updated
What MiniMax H3 does best
MiniMax H3 is built for shots that look finished. We made the clips on this page on Pixel Dojo across six genres: a perfume spot with a title card, a dialogue scene on a night train, an aerial shot over a glacier fjord, a claymation fox baker, a dancer in a shaft of sunlight, and a word built out of pouring sand.
Its standout skill is text. Write a title, a product name or an end slate in capitals and H3 spells it exactly. NOCTURNE 09 landed cleanly on both of our perfume clips, and on the sand prompt H3 built the letters of HORIZON out of the falling grains themselves.
Every clip comes with sound in the same file, shaped by the Sound line of your prompt. At 2K you get full-size masters: our 16:9 files measure 2560 by 1440 and the 21:9 fjord shot measures 2944 by 1248. Vertical 9:16 renders natively, so a vertical cut keeps the whole frame.
H3 also takes direction well. Ask for a locked-off camera with no zoom and no pan and the frame holds still. Ask for a forward drone glide over the ice and you get the glide.
What you can create with MiniMax H3
We made these on Pixel Dojo with MiniMax H3, each from a single text prompt: an epic landscape, a vertical dance piece, a product spot, a dialogue scene and a claymation short. Tap Use this prompt to start from any of them, then match the settings shown on the tile.
A 2K aerial glide over a glacier fjord in 21:9: the red ship carves a white wake between sunlit ice cliffs, and a slab of ice collapses into the sea in a burst of spray. Start from this prompt for a widescreen opener.
Aerial drone shot gliding forward low over a glacier fjord at sunrise. A small red research ship cuts a bright white wake through dark blue water between towering blue ice cliffs, and a slab of ice breaks from the cliff on the left and crashes into the sea. Low golden sun, long shadows, crisp detail on the ice, cinematic teal and amber grade. Sound: steady wind, the distant hum of the ship engine, a deep booming splash as the ice hits the water.
A dancer in crimson silk spins under a single shaft of sunlight, rendered natively in 9:16 for vertical feeds. The skirt flares into a full circle on every turn as one hard beam of light cuts through the empty hall.
Vertical full-body shot, slow push in, symmetrical framing. A dancer in a flowing crimson silk gown spins at the center of an empty white concrete hall as a single hard beam of sunlight cuts down from a high window, the skirt fanning out in a wide circle and dust drifting through the light. 35mm lens, high contrast, deep shadows, clean minimal color grade. Sound: bare feet turning on concrete, the rush of silk, a sparse solo piano.
A 2K product spot with a title card: water runs down a smoky black glass bottle, amber light traces its edges, and the product name NOCTURNE 09 sits beneath it, spelled exactly as written.
Extreme close-up macro, slow push in. A single drop of water falls onto the shoulder of a faceted smoky black glass perfume bottle standing on wet slate and bursts into a crown of fine spray, a thin line of amber backlight tracing the glass edges. At 3 seconds the words NOCTURNE 09 fade in beneath the bottle in thin white serif capitals. 100mm macro lens, shallow depth of field, deep blacks with warm amber highlights, glossy commercial finish. Sound: one crisp water drip, a low sustained cello note, quiet room tone.
A quiet dialogue scene on a night train: she watches the rain, then turns to the lens and speaks under a warm lamp while the camera holds perfectly still. The prompt splits the action into timed beats, a structure worth copying for any performance.
Close-up, locked-off camera, no zoom and no pan. A woman in her forties with short silver hair and a moss green wool coat sits beside a rain-streaked train window at night as streetlights slide across her face. 0 to 2 seconds: she watches the rain. 2 to 5 seconds: she turns her head to face the camera, meets the lens and says softly, "I kept the ticket. I always keep the ticket." Warm practical light from a reading lamp, 50mm lens, fine 35mm film grain. Sound: the steady rhythm of the train on the tracks, rain tapping the glass, her voice close and quiet.
Stop-motion claymation: a fox baker in a striped apron pulls a tray of golden star cookies from a brick oven as steam curls up. Felt curtains, clay texture and warm tungsten light make it read like a real miniature set.
Stop-motion claymation, medium shot, locked-off camera. A small clay fox baker in a blue striped apron pulls a tray of glowing star-shaped cookies out of a tiny brick oven, steam curls up, and the fox does a proud little hop. Visible thumbprints in the clay, felt curtains and a miniature wooden kitchen set, warm tungsten light, gentle stop-motion frame cadence. Sound: a creaking oven door, a cheerful bell ding, a plucked ukulele melody.
How to get results like these from MiniMax H3
Two more clips we made from the same prompts at 768p and 5 seconds, 12.5 credits each, plus the prompt habits behind every example on this page.
The perfume prompt at 768p: a faceted glass bottle catches amber light as spray bursts off its shoulder, and NOCTURNE 09 appears in clean white capitals.
Extreme close-up macro, slow push in. A single drop of water falls onto the shoulder of a faceted smoky black glass perfume bottle standing on wet slate and bursts into a crown of fine spray, a thin line of amber backlight tracing the glass edges. At 3 seconds the words NOCTURNE 09 fade in beneath the bottle in thin white serif capitals. 100mm macro lens, shallow depth of field, deep blacks with warm amber highlights, glossy commercial finish. Sound: one crisp water drip, a low sustained cello note, quiet room tone.
The fjord prompt at 768p: a solid slab of ice breaks from the cliff and crashes into the sea beside the red ship, all in golden sunrise light.
Aerial drone shot gliding forward low over a glacier fjord at sunrise. A small red research ship cuts a bright white wake through dark blue water between towering blue ice cliffs, and a slab of ice breaks from the cliff on the left and crashes into the sea. Low golden sun, long shadows, crisp detail on the ice, cinematic teal and amber grade. Sound: steady wind, the distant hum of the ship engine, a deep booming splash as the ice hits the water.
- Put on-screen words in capitals and say where they sit. NOCTURNE 09 and HORIZON came back spelled exactly as written in every clip we asked for text.
- Write the prompt like a shot note: framing and camera move first, then one subject and one action, then light, lens and a Sound line. Every prompt on this page follows that order.
- Give a performance timed beats and put the spoken line in its own beat. That is how our train scene gets its turn to camera.
- Tell the camera what not to do. "Locked-off camera, no zoom and no pan" kept our train close-up perfectly still.
- Pick the resolution for where the clip will play: 768p at 2.5 credits per second for social feeds, 2K at 4 credits per second for 2560 by 1440 masters and big screens.
- Go beyond text prompts: H3 also starts from a first frame or reference images, and our MiniMax H3 prompting guide walks through both.
MiniMax H3 vs Seedance 2.5 on one sand lettering prompt
We gave both models the same prompt: golden sand pours onto black glass, settles into the word HORIZON, then a gust lifts it. Both spelled the word correctly and both came back with sound. H3 builds the letters out of the pour itself, grain by grain, while Seedance 2.5 starts from finished lettering and throws the whole word into a golden cloud.
Same prompt: Macro shot, slow push in. Fine golden sand pours onto a sheet of black glass and settles into the word HORIZON in bold capital letters, then a gust of wind lifts the letters into a swirling golden cloud. Hard side light, 100mm macro lens, shallow depth of field, deep black background. Sound: hissing sand, a soft wind gust, a low synth swell.
MiniMax H3 at 768p: the pour builds HORIZON letter by letter in raised golden sand, spelled correctly, then the wind starts to lift the middle letters.
Macro shot, slow push in. Fine golden sand pours onto a sheet of black glass and settles into the word HORIZON in bold capital letters, then a gust of wind lifts the letters into a swirling golden cloud. Hard side light, 100mm macro lens, shallow depth of field, deep black background. Sound: hissing sand, a soft wind gust, a low synth swell.
Seedance 2.5 at 720p: HORIZON appears as flat, correctly spelled lettering, then the gust lifts the whole word into a swirling golden cloud.
Macro shot, slow push in. Fine golden sand pours onto a sheet of black glass and settles into the word HORIZON in bold capital letters, then a gust of wind lifts the letters into a swirling golden cloud. Hard side light, 100mm macro lens, shallow depth of field, deep black background. Sound: hissing sand, a soft wind gust, a low synth swell.
| MiniMax H3 | Seedance 2.5 | |
|---|---|---|
| How the word forms | Built letter by letter from the falling sand | Finished flat lettering from the first frame |
| Spelling | HORIZON, correct | HORIZON, correct |
| The gust | Lifts the middle letters while the word still reads | Lifts the whole word into a golden cloud |
| Sound | Stereo track in the file | Stereo track in the file |
| Credits for this clip | 12.5 credits at 768p | 30 credits at 720p |
Best uses for MiniMax H3
Title cards and product names
Launch teasers, product reveals and end slates with the name spelled exactly as you typed it. Write the words in capitals and say where in the frame they sit.
2K masters for big screens
A 16:9 2K clip comes out at 2560 by 1440 and a 21:9 clip at 2944 by 1248, sharp enough for a hero banner or a widescreen opener.
Vertical social video
Pick 9:16 and H3 renders the vertical frame natively, like the dancer above, so nothing is cropped out of a landscape shot.
Dialogue and quiet drama
Give the spoken line its own timed beat and H3 turns the character to camera to speak, with a locked-off frame when you ask for one.
Stylized animation
Claymation, miniatures and other stylized looks keep their character through the motion, like the fox baker and its tray of star cookies.
MiniMax H3 pricing: credits per clip
MiniMax H3 is included in every Pixel Dojo plan, and plans start at $10 a month. It uses credits per second of video: 2.5 credits per second at 768p and 4 credits per second at 2K, for clips from 5 to 15 seconds.
At the settings on this page, a 5 second clip is 12.5 credits at 768p and 20 credits at 2K, our 10 second 2K fjord shot is 40 credits, and a full 15 second clip at 2K is 60 credits.
The same tool page has lower-rate siblings: H3 Max Turbo at 1.25 credits per second at 768p and 0.9 credits per second at 480p, H3 Max at 2 credits per second at 768p, and H3 Fast at 1.5 credits per second at 480p. H3 itself is the one for 2K, and a clip that starts from your own first frame renders at 2K.
MiniMax H3 questions
What is MiniMax H3?
MiniMax H3 is the next-generation AI video model from MiniMax, following Hailuo 01 and Hailuo 02. It makes 5 to 15 second clips at 2K or 768p, with sound, from a text prompt, a first frame or reference images, and it is especially good at putting correctly spelled words on screen.
How do I use MiniMax H3 on Pixel Dojo?
Open the MiniMax H3 tool, write your prompt as a shot note (framing and camera, subject and action, light and lens, then a Sound line), pick 2K or 768p, a length and an aspect ratio, and generate. Every example on this page has a Use this prompt button that starts you from its exact wording.
What resolution and length does MiniMax H3 support?
Clips run 5 to 15 seconds at 2K or 768p in six aspect ratios: 16:9, 9:16, 4:3, 3:4, 1:1 and 21:9. Our 16:9 2K files measure 2560 by 1440 and our 21:9 2K file measures 2944 by 1248.
Does MiniMax H3 generate sound?
Yes. Every H3 clip we made came back with a stereo audio track in the same file. Add a Sound line to your prompt, such as wind, rain on glass, a solo piano or a spoken line, to shape it.
How much does MiniMax H3 cost?
MiniMax H3 is included in every Pixel Dojo plan, from $10 a month. It uses 2.5 credits per second at 768p and 4 credits per second at 2K, so a 5 second clip is 12.5 credits at 768p or 20 credits at 2K.
Is MiniMax H3 the same as Hailuo 3?
MiniMax makes both. Its own announcement presents H3 as its next-generation video model after Hailuo 01 and Hailuo 02, and MiniMax has released the weights as open source. Some sites call H3 Hailuo 3, but MiniMax does not use that name. Hailuo 2.3, the earlier model, has its own page on Pixel Dojo.
Can I use MiniMax H3 through an API?
Yes. Send your prompt to /api/v1/models/minimax-h3/run with a Pixel Dojo API key. The MiniMax H3 API docs page lists every parameter, including duration, resolution, aspect ratio and reference images.
Make your first MiniMax H3 clip
Write a prompt above or start from any example on this page. One subscription covers every model on Pixel Dojo.
Start with MiniMax H3