Video model · MiniMax ·
MiniMax H3 is a video model by MiniMax. In Endless it costs from 84 credits per video and is available on every plan.
MiniMax H3 is MiniMax's omni-modal video model for 5 to 15 second clips at up to 2K with native stereo audio, from text, images or a clip.
MiniMax H3, released in July 2026, models picture and sound together, so dialogue, sound effects and music come out of the same generation as the video, in native stereo. It produces clips of 5 to 15 seconds at 720p or 2K, in aspect ratios from 21:9 to 9:16. As an omni-modal model it reads text, images and video in one prompt.
You can animate a single image as the opening frame, pass several reference images to fix a subject and style, or add a reference clip whose motion and style guide the result. That fits character-led social clips and short brand films where the sound mix matters as much as the picture.
MiniMax H3 Max Lip-sync, also in the Endless catalog, does a different job: it turns a portrait and an audio track into a talking video without a prompt. Use H3 to create a scene from a description, and the lip-sync model when you already have the voice track and the face.
MiniMax H3 generates picture and sound together. MiniMax describes it as an omni-modal model that reads text, images, video and audio in one context, and outputs video with native stereo audio. That changes how you write: the prompt is a timeline of what is seen and what is heard, not only a description of the picture.
MiniMax's own writing guide organizes every prompt into three parts:
On fal, the provider Endless uses for H3, the endpoint rewrites your prompt into this structure before generation by default. You can write in plain sentences and still get a structured prompt underneath. Writing the three parts yourself gives you more control over what ends up in each one.
A short template:
[Style], [shot size]. [Subject and setting]. [Action, in order].
[Camera movement]. At 00:04.000 the camera cuts to [next shot].
Soundscape: [ambience and physical sounds].
Music: [instruments and tempo], or no music.
MiniMax recommends opening with the overall style and the starting composition, for example "Live-action, cinematic, a medium-wide shot of..." Other styles named in the guide include 2D animation, 3D CG, claymation, watercolor and vintage film.
For more than one shot, give each cut a time inside the clip: "At 00:04.000 the camera cuts to a close-up of...". Use a single timestamp, not a range, and keep the times increasing. A cut should bring new information: a new subject, place, state, viewpoint or moment. If you only need to get closer or shift the angle a little, use a camera movement instead.
The guide describes camera motion with three parts: the type of movement, its amplitude and its speed. Types include zoom in and out, push in and pull out, pan, truck left and right, tilt, pedestal up and down, arc shot, tracking shot, static shot, POV, slight or strong shake, and roll. Add "with small amplitude" or "with large amplitude", and "at slow speed" or "at fast speed", only when they matter. Write it as a sentence: "The camera pushes in with small amplitude at slow speed toward the letter in her hands."
Introduce each speaker with enough detail to define the voice: age, gender, pitch, timbre, speed or accent. Then give the line. Keep the same description for a speaker across shots. MiniMax lists stable dialogue support for 11 languages: Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian and Spanish. Write each line in the language it should be spoken in.
For narration, say the character speaks "in an off-screen voiceover" and that their lips stay closed, so the model does not animate their mouth. Signs, labels and other on-screen text go in double quotes, word for word.
H3 in Endless accepts three kinds of input besides the prompt:
In your workspace, open Video and pick MiniMax H3 in the model picker.
Write a prompt with the subject, camera and motion, or start from an image you upload or generate.
Generate, then refine in chat or on a canvas. Every clip is saved to your workspace assets.
Grouped by use case. Each one opens in Endless with MiniMax H3, ready to edit and generate.
Live-action, cinematic. A medium-wide shot of a small roastery at dawn as a roaster tips fresh beans into a steel cooling tray. The camera pushes in slowly as the beans crackle and smoke curls up through a shaft of light. Soundscape: crackling beans, a fan humming, a metal scoop scraping. Music: a slow acoustic guitar pattern that fades at the end.
Cinematic. A cyclist rides along a coastal road at sunrise, tracked from the side at fast speed. At 00:04.000 the camera cuts to a close-up of the chain and gears turning. At 00:07.000 it cuts to a wide shot as she crests the hill and the sea opens up below. Wind rushing, tires on asphalt, gulls. Music: steady electronic pulse building in volume.
A static close-up of a ceramic mug on a wooden table as hot tea is poured from a cast-iron kettle. Steam rises and catches window light. A wooden sign behind it reads "OPEN 7 AM". Soundscape: pouring water, a soft clink, rain on the window. No music.
3D CG, bright and clean. A pair of wireless earbuds spins slowly above a white pedestal, then the case snaps shut and the camera arcs around it at slow speed. Soundscape: a soft magnetic click and a low whoosh. Music: minimal synth notes at a moderate tempo.
Live-action, warm evening light. A baker in a flour-dusted apron pulls a tray of croissants from a stone oven. The camera trucks right with small amplitude as she sets the tray down and smiles. Soundscape: oven door creaking, tray clanking, faint street noise. Music: upright bass and brushed drums, light and slow.
Live-action. Two neighbors talk over a garden fence on a sunny afternoon. The older man with a slow, gravelly voice holds up a tomato and says: 'Biggest one this year.' The young woman with a quick, bright voice laughs and answers: 'You say that every year.' Soundscape: bees, distant lawnmower, leaves rustling.
Cinematic. A lighthouse keeper sits at a desk writing in a logbook by lamplight. He narrates in an off-screen voiceover, with his lips closed: 'Third night of fog. No ships tonight.' The camera pushes in slowly toward the page. Soundscape: waves against rocks, a foghorn far away, pen scratching.
A street vendor in Mexico City hands a customer a cup of elote. The vendor, cheerful and loud, says in Spanish: '¡Con todo, como te gusta!' The customer answers in English, smiling: 'Perfect, thank you.' Soundscape: traffic, music from a nearby stall, chatter.
Inside a car parked in the rain at night. The driver, a woman with a soft tired voice, looks at the passenger and says: 'We can still turn around.' The passenger, a man with a low calm voice, keeps his eyes on the road ahead and says: 'No. We keep going.' Soundscape: rain on the roof, wipers, a low engine idle.
A teacher stands at a chalkboard in a sunlit classroom and turns to the class. In a warm, clear voice she says in French: 'Bonjour à tous, on commence.' The camera cuts at 00:03.000 to the students opening their notebooks. Soundscape: chairs scraping, pages turning, birds outside.
Image 1 is the main character, a woman in a yellow raincoat. Image 2 is her small gray dog. Keep both consistent with their references as they walk together across a wet cobblestone square at dusk, the dog shaking off rain. Soundscape: light rain, paws on stone, distant church bells.
Image 1 is the hero's costume and face. Image 2 is the desert canyon setting. The hero walks into the canyon at midday, stops, and shades her eyes to look at the horizon. The camera pedestals up slowly to reveal the canyon walls. Wind and loose stones falling.
Image 1 is the mascot character. Keep its colors and proportions while it waves at the camera in front of a toy store window, then jumps and lands with a small bounce. Soundscape: a squeaky step and children laughing nearby. Music: playful xylophone melody.
Image 1 is the chef. Image 2 is the restaurant interior. The chef walks through the dining room carrying a plate, sets it down at a table, and says with a proud, warm voice: 'Tonight's special.' The camera tracks beside him. Soundscape: soft chatter, cutlery clinking.
Image 1 defines the visual style, a watercolor illustration. Image 2 is the fox character. In that watercolor style, the fox trots through a snowy forest, stops to sniff the air, and leaps over a fallen log. Soundscape: snow crunching, wind in the pines.
The woman in the picture lifts her gaze from the book toward the window as the afternoon light shifts across her face. The camera pushes in with small amplitude at slow speed. Her clothing, chair and the room stay as they are. Soundscape: a clock ticking, pages turning.
Starting from this street photo, the shop lights switch on one by one as evening falls. A tram passes from left to right and pedestrians begin to move. The camera holds a static shot. Soundscape: tram bell, footsteps, distant traffic.
The dog in the picture tilts its head, then bolts toward the camera across the grass, ears flapping. The camera pulls out at fast speed to keep it in frame. Soundscape: panting, grass rustling, a short bark.
Starting from this product shot, the bottle cap twists open on its own and a thin mist sprays upward, catching the light. The camera arcs a quarter turn around the bottle. The label text stays exactly as in the image. Soundscape: a soft hiss and a click.
From this mountain landscape, clouds race over the peaks and shadows sweep down the valley as the light moves from morning to dusk. The camera tilts down slowly toward the river. Soundscape: wind, a distant eagle call. Music: sustained low strings.
Video 1 provides the dance movement. Image 1 is the dancer. The dancer from Image 1 performs the choreography from Video 1 on a rooftop at sunset, the city skyline behind her. Soundscape: wind and the city below. Music: upbeat electronic track.
Video 1 provides the camera movement and pacing. Recreate that same camera path through a quiet forest at dawn, with fog between the trees and light breaking through. Soundscape: birdsong and dripping leaves.
Video 1 is the motion reference for a runner. Image 1 is the athlete. The athlete from Image 1 runs with the stride and timing of Video 1 along a beach at low tide. Soundscape: waves, breathing, feet slapping wet sand.
Video 1 provides the gesture timing. Image 1 is a cartoon robot. The robot repeats the waving and bowing from Video 1 in front of a workshop bench full of tools. Soundscape: small servo whirs and a cheerful beep.
Video 1 sets the rhythm of the cuts. Image 1 is the sneaker. Build a short product sequence with the same cut rhythm: the sneaker on concrete, a close-up of the laces, a wide shot of a skater wearing it. Music: punchy drum loop.
Ultra-wide 21:9, cinematic. A lone rider on horseback crosses a vast salt flat at dusk, small in the frame. The camera trucks left slowly, keeping the rider centered as the sky turns pink. Soundscape: hooves on crust, wind. Music: a slow solo cello.
Vertical 9:16, live-action. A street dancer spins on a basketball court at night under a single floodlight. The camera tilts up from her sneakers to her face as she finishes the move and grins. Soundscape: sneakers squeaking, a bouncing ball nearby. Music: hip-hop beat.
Vertical 9:16. A barista slides a latte across the counter toward the camera and says in a friendly voice: 'Your usual.' The camera pushes in on the cup. Soundscape: espresso machine hiss, café chatter.
Ultra-wide 21:9. A cargo ship moves through a narrow fjord at sunrise, cliffs on both sides. The camera pedestals up with large amplitude to reveal the full length of the fjord. Soundscape: engine drone, water against the hull, gulls.
Square 1:1. Top-down static shot of hands arranging fruit on a wooden board: figs, grapes, a sliced orange. Each piece lands with a soft tap. Soundscape: soft taps and a knife slicing. No music.
by ByteDance
30-second clips in one pass, guided by up to 50 references
From 51 credits per video
by ByteDance
Clips of 4 to 15 seconds steered by image, video and audio references
From 16 credits per video
by Black Forest Labs
Clips up to 20 seconds with sound, from a prompt, frames or a clip
From 90 credits per video
MiniMax H3 next to other models: price, maker and plan side by side.
Start free, create a project, and bring your brand’s references together. Or book a demo to explore Endless with your team.