People who speak
Put a line in double quotes and describe the voice. The person says it, and the room around them sounds right.
MiniMax H3 Turbo Prompting Guide
MiniMax H3 Turbo is the fast, low-cost version of MiniMax H3. It makes 5 to 15 second clips with sound, from a prompt, a photo, or up to 9 reference images. A 5 second clip at 720p is 7.5 credits. This guide shows the prompts that work, with the clip each one made.
Every H3 Turbo clip comes with sound: voices, footsteps, waves, room tone. You write the sound into the prompt the same way you write the picture, and people can speak the lines you put in quotes.
It costs 1.5 credits per second at 720p and 0.75 credits per second at 480p. That makes it the place to try an idea a few times before you spend more on a longer or sharper clip.
Every clip on this page was made with the prompt shown beside it, in one try, on Pixel Dojo.
5 to 15 s
Clip length
1.5 credits
Per second at 720p
9
Reference images
Put a line in double quotes and describe the voice. The person says it, and the room around them sounds right.
Add a photo and the clip opens on it. Say what happens next: a turn, a walk, a smile.
Add a photo of each person, product or place and name them in the prompt. They show up together in a scene you describe.
Pick the shape before you generate. 9:16 fills a phone screen for Reels, Shorts and TikTok.
Each example shows the exact prompt that produced the result. Copy any prompt with one click.
720p · 16:9 · 5 s · 7.5 credits
A beautiful woman in her late twenties with long sun-streaked blonde hair, bright blue eyes and a warm easy smile, in a white linen shirt, sits on the bow of a small wooden boat at golden hour and tucks her hair behind her ear as the wind lifts it. Low warm sun from the right, sparkling turquoise water softly blurred behind her, natural skin texture, shallow depth of field. Begin on a medium close-up at eye level, then slowly push in as she looks into the lens and says in a relaxed American voice: "Okay. One more take, and then we swim." Audio: her clear close voice, water lapping against the hull, a light breeze, distant gulls. No background music, no subtitles.
Describe the person, give one small action, then put the line in double quotes and say how the voice sounds. Close the prompt with the sounds you want and "no background music, no subtitles".
720p · 16:9 · 10 s · 15 credits
A stylish couple in their late twenties at a small marble table outside a Paris cafe in the morning. She has a sleek dark bob and red lipstick and wears a camel coat. He has tousled brown hair and a charcoal sweater. Soft overcast light, warm cafe windows behind them, shallow depth of field, natural skin texture. 0-5 s: medium two-shot, steady camera, as she stirs her espresso, looks up at him and says in a playful voice: "You said five minutes. That was an hour ago." 5-10 s: slow push in on him as he laughs, raises both hands and answers in a warm low voice: "The croissants were worth it. Admit it." One continuous shot, no cuts. Audio: their two voices one at a time, a spoon on porcelain, quiet street ambience. No background music, no subtitles.
Split a longer clip into time ranges and give each range one speaker. "One continuous shot, no cuts" keeps it as a single take, and "one at a time" stops the voices from overlapping.
720p · 16:9 · 5 s · 7.5 credits
A handsome athletic man in his late twenties with wet dark curly hair and a short beard, in a black wetsuit, carves a hard turn on a white surfboard along the face of a glassy green wave at sunrise, spray fanning off the rail. Backlit by a low golden sun, the spray glowing, the water translucent and deep green. Begin on a tracking shot at water level beside him, then tilt up as the lip of the wave curls over his shoulder. Audio: the rush and hiss of the wave, the board cutting water, a short breath, wind. No dialogue, no background music.
Say where the light comes from and what it hits. "Backlit by a low golden sun, the spray glowing" does more than any quality word. Give the camera one path to follow.
720p · 16:9 · 5 s · 7.5 credits
A faceted amber glass perfume bottle with a gold cap stands on a slab of wet black stone as a slow wave of clear water sweeps across the stone and splashes up around the base, droplets hanging in the air. One hard rim light from the right, deep teal gradient background, crisp reflections, macro detail on the glass. Begin on a close-up at table height, then slowly orbit a quarter turn to the left as the splash settles. Audio: a deep soft whoosh, the splash, droplets pattering on stone, then silence. No dialogue, no background music.
One object, one event, one camera move. Name the surface and the single light, and describe the sound from start to finish so the clip ends clean.
720p · 16:9 · 5 s · 7.5 credits
Stop-motion claymation: a cheerful round clay baker with a tall white hat and rosy cheeks slides a tray of tiny clay bread loaves out of a glowing brick oven and blows on them, little puffs of clay steam rising. Warm orange oven glow, a cozy miniature kitchen with handmade textures and visible fingerprints in the clay. Begin on a wide shot of the kitchen, then push in to the baker and the steaming tray. Audio: a crackling oven, a wooden tray sliding, a happy little hum, a soft puff of breath. No dialogue, no background music.
Lead with the style, then describe the materials ("visible fingerprints in the clay"). The sound list still applies: small, close sounds suit a miniature set.
720p · 9:16 · 5 s · 7.5 credits
A beautiful dancer in her twenties with long dark hair in a high ponytail, in a flowing red dress, spins on a rooftop at sunset with the city skyline behind her, the dress and her hair flaring out as she turns and lands facing the camera with a confident smile. Warm orange backlight, long shadows, soft lens flare, natural skin texture. Begin on a full-length shot, then slowly push in to a medium shot as she finishes the spin. Audio: the swish of fabric, her shoes on concrete, distant city hum, a light wind. No dialogue, no background music.
Pick 9:16, start full length and push in. A move with a clear end ("lands facing the camera") gives the clip a finish instead of a fade.
720p · start photo · 5 s · 7.5 credits
The woman in <Picture 1> continues down the marble staircase, one hand lifting the green gown as the pleats ripple, then stops, turns her head toward the camera and gives a small confident smile. The hard shaft of daylight from the left stays the same, dust drifting in the beam. Begin on the framing of the photo, then slowly push in. Audio: her heels on marble, the soft rustle of the gown, a quiet echo in the hall. No dialogue, no background music.
Add a start photo and call it <Picture 1>. Say what happens next and what stays the same. The clip keeps the shape of your photo.
720p · 16:9 · 5 s · 7.5 credits · two reference images
The woman in <Picture 1> and the man in <Picture 2> walk side by side through a night market under strings of warm lanterns, sharing a paper cone of roasted chestnuts and laughing. Keep both faces and hair exactly as in the pictures. She wears a cream sweater, he wears a navy sweater. Warm lantern light, steam from food stalls, soft blurred crowd behind them, shallow depth of field. Begin on a medium shot tracking backward in front of them at walking pace. Audio: market chatter, sizzling food, their laughter, footsteps. No background music, no subtitles.
Add each person as a reference image and name them in order: <Picture 1>, <Picture 2>. Say "keep both faces and hair exactly as in the pictures", then describe the new scene and outfits.
Who does what and where. The light and materials. "Begin on" a shot, "then" one camera move. "Audio:" the sounds in the order they happen.
Name the opening shot and a single move, like a slow push in or a track beside the subject. Leave it out and the model picks its own shots, which can mean a cut you did not want.
Put spoken words in double quotes and add how they sound: "in a relaxed American voice". With two people, give each line its own time range.
End the sound list with "no background music, no subtitles" when you want a clean clip you can score or caption yourself.
A start photo is <Picture 1>. Reference images are numbered in the order you add them. Tell the model what each one is: "the woman in <Picture 1>".
480p is half the price. Use it to check the action and the camera move, then run the one you like at 720p.
For 10 or 15 seconds, write "0-5 s:" and "5-10 s:" and give each range one event. Add "one continuous shot, no cuts" for a single take.
| Setting | Values | Notes |
|---|---|---|
| Length | 5 to 15 seconds | Price scales with length. |
| Resolution | 480p · 720p | 0.75 credits per second at 480p, 1.5 at 720p. |
| Shape | 16:9 · 4:3 · 1:1 · 9:16 · 3:4 · 21:9 | A start photo sets the shape itself. |
| Sound | Always on | Describe it in the prompt. |
| Start and end photo | Optional | The clip opens on the start photo and can finish on an end photo. |
| References | 9 images · 3 videos · 3 audio clips | Name them <Picture 1>, <Video 1>, <Audio 1>. Audio needs an image or video beside it. |
MiniMax H3 Turbo is the fast, low-cost version of the MiniMax H3 video model. It makes 5 to 15 second clips at 480p or 720p, always with sound, from a prompt, a start photo, or reference images, videos and audio.
1.5 credits per second at 720p and 0.75 credits per second at 480p. A 5 second clip at 720p is 7.5 credits and a 15 second clip is 22.5. Every Pixel Dojo plan includes it.
H3 Turbo is the lower-cost one, at 480p and 720p. MiniMax H3 goes up to 2K. Many people try a shot on H3 Turbo first, then make the final on MiniMax H3 when they need the extra sharpness.
Yes. Add a start photo and the clip opens on it. Call it <Picture 1> in the prompt and describe what happens next. You can also add an end photo for the clip to finish on.
Yes. Put the line in double quotes and describe the voice. For two speakers, give each line its own time range and write "one at a time".
Add up to 9 images of people, products or places. They are numbered in the order you add them, so write "the woman in <Picture 1> and the man in <Picture 2>" and then describe the scene.
Yes. It is in the Canvas model picker as MiniMax H3 Turbo, and the API endpoint is /api/v1/models/minimax-h3-turbo/run with prompt, duration, resolution, aspect_ratio, image_url and reference_images.