Prompt to actual result

MiniMax H3 video examples

Watch every usable output from the reviewed H3 manual beside an English prompt translation. Filter by the skill you want to learn, then open any lesson for the complete prompt, control pattern, and failure check.

64 official lessons
What is included

Generated films, editing tasks, motion design, AR, product work, game UI, animation, character, and audio lessons. Reference-input clips are excluded. Videos stream from Cloudflare R2 only after you press play.

Open the H3 manual
Browse by lesson type

Find the result you want to reproduce.

64 results
Desert luxury campaign MiniMax H3 result frame Actual H3 output / 15s
Brand film / Reference Generation

Desert luxury campaign

The 15-second result keeps the twilight road, two performers, leather wardrobe, car, bag reveal, and campaign typography inside one coherent fashion edit.

English prompt translation

Create a premium 16:9 fashion campaign on a desert highway beside a vintage car. Image 1 defines the atmosphere, location, and film texture. Image 2 locks the performer. Image 3 locks the black leather bag. Ima...

Telescope fashion keyframes MiniMax H3 result frame Actual H3 output / 15s
Fashion transition / Reference Generation

Telescope fashion keyframes

The result holds the binocular mask while moving through architecture, fabric, a street-fashion figure, and eyewear with blur-timed transitions.

English prompt translation

Use Images 1 through 4 as consecutive keyframes seen through an old binocular viewfinder searching for a MINIMAX installation. Begin out of focus with restrained handheld movement, then push in quickly and rack...

Space opera title trailer MiniMax H3 result frame Actual H3 output / 15s
Film trailer / Reference Generation

Space opera title trailer

The finished trailer combines a fleet departure, lead-character close-up, title cards, hard editorial beats, and synchronized cinematic sound.

English prompt translation

Create a realistic cinematic space-opera trailer with high-contrast lighting and a tight pace. Image 1 defines the atmosphere and visual style; Image 2 locks the lead character. Shot 1 is an ultra-wide establis...

Yellow motion-graphics city MiniMax H3 result frame Actual H3 output / 15s
Motion graphics / Text-to-Video

Yellow motion-graphics city

The output maintains a yellow, black, and white poster system across cel-animation action, geometric wipes, flat typography, and graphic transitions.

English prompt translation

Pure 2D cel animation combined with abstract motion graphics, designed like a continuously moving flat poster. Build a high-contrast yellow visual system using gold, amber, lemon yellow, black, and white, with ...

Visual novel interface transition MiniMax H3 result frame Actual H3 output / 15s
Game interface / First/Last-Frame Image-to-Video

Visual novel interface transition

The official result preserves a romance-game interface while moving from a player choice into a controlled backstage character response.

English prompt translation

Use Image 1 as the opening frame and Image 2 as the strict final frame. Create a premium Chinese romance visual-novel interface transition that moves from the choice “Go watch his performance” to a backstage mo...

Coffee-to-desert match transition MiniMax H3 result frame Actual H3 output / 15s
Material match cut / First/Last-Frame Image-to-Video

Coffee-to-desert match transition

The camera pushes through coffee foam and cocoa texture until the same granular motion resolves into a wide desert landscape without a hard cut.

English prompt translation

Begin from Image 1. Push quickly into the milk foam, cocoa particles, and dark coffee texture until the bubbles, powder, and liquid ripples fill the screen and hide every other object. Maintain realistic macro ...

Selective voxel reality edit MiniMax H3 result frame Actual H3 output / 15s
Video editing / Reference Generation

Selective voxel reality edit

Buildings and pedestrians remain live action while selected cars and trees become light-matched voxel objects inside the original moving shot.

English prompt translation

Keep the buildings, pedestrians, camera motion, and overall environment from Video 1 in a realistic live-action style. Change only the trees and cars into 3D pixel or voxel building blocks, using Image 1 as the...

Reference-action foam fight MiniMax H3 result frame Actual H3 output / 15s
Performance transfer / Reference Generation

Reference-action foam fight

The generated characters reproduce the referenced plate handoff, surprise foam throw, reaction, and playful escalation inside one coherent kitchen scene.

English prompt translation

Image 1 locks the characters and their appearance. Video 1 strictly defines their actions, facial performance, and timing. At a sink on the right side of frame, the man hands a washed plate to the woman at the ...

Sci-fi mystery trailer MiniMax H3 result frame Actual H3 output / 15s
Film trailer / Reference Generation

Sci-fi mystery trailer

The result uses a monumental gate, small human silhouette, wet reflections, rust-red typography, and trailer-style sound timing.

English prompt translation

Create a realistic, high-contrast sci-fi mystery trailer. Image 1 defines the atmosphere and Image 2 locks the protagonist. Open on an enormous circular cosmic gate over a wet reflective floor, with the charact...

Fire and VHS fashion campaign MiniMax H3 result frame Actual H3 output / 15s
Fashion film / Reference Generation

Fire and VHS fashion campaign

The output preserves the performer and black wardrobe across fire-lit close-ups, analog faults, and fast fashion cuts.

English prompt translation

Use Image 1 for the overall mood and Image 2 for the performer. Create a 15-second 16:9 fashion film with platinum hair, narrow black sunglasses, a glossy black patent coat, and cool confidence. Mix night fire,...

Glowing kitchen doodle MiniMax H3 result frame Actual H3 output / 15s
Hybrid live action / Text-to-Video

Glowing kitchen doodle

A handheld kitchen scene becomes a warm, plausible encounter with small animated light forms.

English prompt translation

Create a 15-second horizontal video that combines a real dusk kitchen with hand-drawn glowing animation. Preserve an improvised phone-camera feel: slight shake, hesitant close focus, exposure fluctuation, and c...

Midnight laundromat doodles MiniMax H3 result frame Actual H3 output / 15s
Hybrid live action / Text-to-Video

Midnight laundromat doodles

The output combines a quiet laundromat, unstable phone capture, and luminous animated interventions.

English prompt translation

Create a 15-second live-action night laundromat mixed with glowing hand-drawn animation. Show fluorescent flicker, running washers, plastic baskets, an old bench, and one lost sock. Use obvious one-handed phone...

Cyber-grunge rap montage MiniMax H3 result frame Actual H3 output / 15s
Music video / Reference Generation

Cyber-grunge rap montage

The finished clip combines fashion portraiture, coarse print texture, hard cuts, and aggressive oversized type.

English prompt translation

Create a dark-pop, cyber-grunge rap video with realistic high-fashion imagery and late-1990s independent magazine texture. Use photocopy marks, film scans, underground poster collage, coarse grain, halftone dot...

Retro crime title sequence MiniMax H3 result frame Actual H3 output / 15s
Title design / Reference Generation

Retro crime title sequence

The result delivers a dense but coherent collage of silhouettes, credits, record motifs, split frames, and crime-jazz timing.

English prompt translation

Create a 15-second light-mystery crime title sequence inspired by retro Japanese animation, hard-edged silhouettes, comic collage, asymmetrical split screens, geometric color fields, English credits, and restra...

Fairy-tale greenscreen composite A MiniMax H3 result frame Actual H3 output / 15s
Scene replacement / Reference Generation

Fairy-tale greenscreen composite A

The performer is integrated into a fantasy environment with matched motion, scale, and scene lighting.

English prompt translation

Remove the green background from Video 1 and replace it with a fairy-tale environment in the style of Video 2. Make every background element respond correctly to the performer action. Relight the performer so d...

Red-white dynamic poster MiniMax H3 result frame Actual H3 output / 10s
Motion poster / First/Last-Frame Image-to-Video

Red-white dynamic poster

The result keeps the vertical poster grid and collectible figure intact while bringing type and layout beats to life.

English prompt translation

Animate the supplied poster while preserving its gallery-style white outer frame, inner frame, red-white-black palette, 3D figurine quality, and exact layout. Introduce the typography with lively motion and sma...

Seaside doodle market A MiniMax H3 result frame Actual H3 output / 15s
Mixed-media sequence / Text-to-Video

Seaside doodle market A

A hand points across a summer market and selectively redraws individual objects without converting the full scene.

English prompt translation

Create a 10-15 second, four-shot phone video at a bright seaside market. In every shot, a correctly held cyan pen enters from below and points to exactly one real object. Only that object becomes rough hand-dra...

Seaside doodle market B MiniMax H3 result frame Actual H3 output / 15s
Mixed-media variant / Text-to-Video

Seaside doodle market B

This alternate output applies the same pen-triggered drawing rule to a different sequence of market objects.

English prompt translation

At a sunlit seaside market, show a cyan pen held naturally from the bottom of frame. Across four quick shots, let it point to one real-world object at a time and convert only that target into loose 2D pencil an...

TouchDesigner tracking overlays MiniMax H3 result frame Actual H3 output / 15s
Video effects / Reference Generation

TouchDesigner tracking overlays

Clean red and cyan tracking rectangles follow movement while the base footage remains visually intact.

English prompt translation

Use Video 1 as the only base and preserve its people, faces, clothing, action, scene, color, camera motion, duration, and frame rate. Add only TouchDesigner-style tracking graphics. Every overlay must be a simp...

H3 continuous-line brand film MiniMax H3 result frame Actual H3 output / 15s
Motion graphics / Reference Generation

H3 continuous-line brand film

The output develops the H3 mark into a continuous waveform and uses one line to connect multimodal concepts.

English prompt translation

Preserve the opening electric-blue MiniMax H3 frame and use Image 2 as the exact target waveform logo. Create a 15-second abstract technology brand film about text, image, video, and audio becoming one shared c...

One becomes everyone MiniMax H3 result frame Actual H3 output / 15s
Brand motion / Text-to-Video

One becomes everyone

Dots, rings, H-shaped links, and three-track forms grow into a compact MiniMax H3 identity system.

English prompt translation

Create a 15-second premium flat-motion brand film for MiniMax H3. The only readable text is "MiniMax H3" and "Intelligence With Everyone." Build the concept "One becomes everyone" from three primitives: a tiny ...

Chinese kinetic typography MiniMax H3 result frame Actual H3 output / 15s
Type experiment / Text-to-Video

Chinese kinetic typography

The finished sequence uses Chinese characters alone to create aggressive scale changes, collisions, repetition, and rhythmic hierarchy.

English prompt translation

Create a Chinese kinetic-type experiment on a completely black background. Allow only white Chinese characters with a very small amount of dark-red offset or emphasis. Do not show people, scenery, photos, illus...

AR city world edit MiniMax H3 result frame Actual H3 output / 14s
AR enhancement / Reference Generation

AR city world edit

A real city intersection is edited in place with a large spatial AR transformation that follows the source camera.

English prompt translation

Use the uploaded source video as the only spatial base. Preserve the real camera motion, hand and mouse positions, intersection geometry, crosswalk, buildings, traffic, trees, signals, perspective, and daylight...

Cat and drill AR comedy MiniMax H3 result frame Actual H3 output / 8s
Reality enhancement / Reference Generation

Cat and drill AR comedy

The original vertical cat clip gains a playful visual effect while its subject and room layout remain stable.

English prompt translation

Enhance Video 1 while preserving its composition, camera angle, cat position and timing, drill, and computer placement. Add exaggerated but physically grounded comedy effects. Do not change the cat appearance, ...

Poster-world AR character MiniMax H3 result frame Actual H3 output / 36s
AR sequence / Reference Generation

Poster-world AR character

The character moves through several flat poster spaces, interacts with typography and a play control, then returns to the final card.

English prompt translation

Have the subject climb out of a white rectangular frame, launch into Image 1, sit on the white dimensional title, and look toward a background person. The subject then throws a thin white elastic line, swings i...

Vertical family confrontation MiniMax H3 result frame Actual H3 output / 15s
Short drama / Text-to-Video

Vertical family confrontation

The vertical result uses warm interiors, close performances, and escalating reverse angles to create a short-drama hook.

English prompt translation

Create a vertical live-action family argument in a Chinese home or small restaurant with warm light, red decoration, calligraphy in the background, shallow depth of field, and tight pacing. Qin Haoxuan responds...

Vampire romance hook MiniMax H3 result frame Actual H3 output / 15s
Vertical trailer / Reference Generation

Vampire romance hook

The output compresses castle atmosphere, identity continuity, and a dangerous romantic confrontation into a vertical teaser.

English prompt translation

Create a 15-second 9:16 live-action vampire romance trailer. Image 1 locks both leads and Image 2 defines the castle. A human heroine enters a forbidden chamber and awakens an aristocratic vampire who senses a ...

Snowy bamboo mystery MiniMax H3 result frame Actual H3 output / 15s
Costume drama / Text-to-Video

Snowy bamboo mystery

Cold fog, snow, bamboo layers, and close character exchanges create a controlled wuxia mystery scene.

English prompt translation

Create a 16:9 historical martial-arts mystery at night in a bamboo forest. Use cold blue, ink green, gray-black, thin fog, drifting snow, foreground leaves, distant white haze, and shallow depth of field. Stage...

White-studio eyewear fashion MiniMax H3 result frame Actual H3 output / 15s
Product campaign / Reference Generation

White-studio eyewear fashion

Two models move through a bright white fashion set while futuristic eyewear remains the visual focus.

English prompt translation

Create a premium 9:16 eyewear campaign with the pacing and white-cyclorama look of the reference video. Image 1 defines the full-body poses and wardrobe of two female models, Image 2 refines their appearance, a...

Ergonomic chair explainer MiniMax H3 result frame Actual H3 output / 15s
Product function / Reference Generation

Ergonomic chair explainer

The vertical product film moves from clean chair beauty shots into mesh, adjustment, posture, and support demonstrations.

English prompt translation

Create a premium 360-degree presentation of the black ergonomic chair from the reference. Show breathable mesh with visible airflow, lumbar support and body-curve engineering animation, armrest and height adjus...

Peking duck product launch MiniMax H3 result frame Actual H3 output / 15s
Food campaign / Text-to-Video

Peking duck product launch

Glossy skin, slow macro rotation, black space, and restrained launch titles turn the dish into a premium product reveal.

English prompt translation

Create a 15-second 16:9 premium product-launch film for a perfectly roasted Peking duck. Use a pure black background, studio lighting, extremely slow rotating macro detail, large minimal sans-serif titles, gene...

Cyberpunk loadout UI MiniMax H3 result frame Actual H3 output / 15s
Game interface / Reference Generation

Cyberpunk loadout UI

The result moves from menu selection to prosthetic customization and then loads a neon third-person game world.

English prompt translation

Use Image 1 for the character and Image 2 for UI style. Begin on a purple menu, choose CONTINUE, then zoom into right-arm equipment and cycle from PHANTOM GRIP to CHRONOS CLAW. Move through armament customizati...

Kinetic product landing page MiniMax H3 result frame Actual H3 output / 15s
Web motion / First/Last-Frame Image-to-Video

Kinetic product landing page

The finished website demo combines bold editorial type, product close-ups, controlled scrolling, and high-impact hover states.

English prompt translation

Animate a forceful product landing page around the supplied hero image. Use huge slanted sans-serif type, speed-driven light, dark carbon-fiber or sports-mesh texture, and a tight rhythmic edit. Demonstrate a s...

Automotive website hero motion MiniMax H3 result frame Actual H3 output / 10s
Web motion / First/Last-Frame Image-to-Video

Automotive website hero motion

The result reveals a black performance car through staged title, information-panel, and red-light animation.

English prompt translation

Animate the supplied automotive website hero. Slide the top title downward into place, bring the lower text panel upward, and gradually change the car lights from dark to red. Preserve the page layout, vehicle ...

Hand-dodge beagle game MiniMax H3 result frame Actual H3 output / 10s
Game concept / Reference Generation

Hand-dodge beagle game

The clip reads like a complete reaction-game loop with clear directional prompts and elastic dodge poses.

English prompt translation

Create a fixed-camera 4:3 hand-drawn reaction game on warm paper texture. The same rounded cartoon beagle responds to highlighted up, down, left, and right arrows by stretching away from a hand entering from th...

Digital pet index UI MiniMax H3 result frame Actual H3 output / 15s
Interface showcase / Reference Generation

Digital pet index UI

The result cycles through luminous digital pets while the catalog layout and cursor remain stable.

English prompt translation

Use Image 1 as the exact UI, layout, and cursor reference, and map Images 2-5 to the named pet cards. Keep the camera fixed and preserve the panels, type hierarchy, icons, rounded card grid, and elegant composi...

Robot arm pick and place MiniMax H3 result frame Actual H3 output / 7s
Industrial demo / First/Last-Frame Image-to-Video

Robot arm pick and place

A clean fixed shot shows the complete robot-arm grasp, transfer, and release sequence.

English prompt translation

Use a fixed camera. The robot arm lowers, grips the red cube, moves it forward, and releases it. Preserve the table, surrounding objects, arm geometry, and the complete physical order of the pick-and-place acti...

Robot arm office background edit MiniMax H3 result frame Actual H3 output / 7s
Industrial edit / Reference Generation

Robot arm office background edit

The same pick-and-place action is retained while the tabletop and background become a coherent office workspace.

English prompt translation

Replace the source table with a standard office workstation and replace the background with blinds and metal filing cabinets. Match desk perspective, scale, highlights, shadows, focus, and depth to the original...

Xianxia character sequence MiniMax H3 result frame Actual H3 output / 15s
Character PV / Reference Generation

Xianxia character sequence

The output carries one detailed fantasy character through dramatic close-ups, back views, and large-scale xianxia action.

English prompt translation

Use Image 2 as the fixed character identity and preserve the black half-up hair, silver openwork crown, dark-blue ribbon, layered pale robes, translucent blue outer robe, deep-blue belt, silver floral clasp, an...

Otome male-lead PV MiniMax H3 result frame Actual H3 output / 15s
Character promo / Reference Generation

Otome male-lead PV

The finished PV presents one consistent male lead through polished romance-game close-ups and stylized transitions.

English prompt translation

Create an otome-game male-lead character PV. Use Image 2 as the strict identity reference and preserve the same face, hairstyle, body proportions, clothing design, material details, and premium romance-game CG ...

Red-room otome scene MiniMax H3 result frame Actual H3 output / 15s
Romance scene / Reference Generation

Red-room otome scene

The result centers on a composed male lead, partial female presence, red velvet, whisky detail, and low-key romantic tension.

English prompt translation

Use Image 1 for the male lead and Image 2 for a dark red-and-black luxury interior. Frame only upper bodies in realistic cinematic close coverage. The woman in a red backless dress remains mostly outside full f...

First-person FPS gameplay MiniMax H3 result frame Actual H3 output / 15s
Game footage / Text-to-Video

First-person FPS gameplay

The clip reads as a short player-controlled advance with scanning, firing, recoil, and continued movement.

English prompt translation

Simulate ordinary first-person gameplay in a modern military FPS. The player advances slowly along cover outside a base, scans the passage, pauses to fire a few rounds at a distant objective, and continues forw...

Clay fox canyon jump MiniMax H3 result frame Actual H3 output / 10s
Stylized action / Text-to-Video

Clay fox canyon jump

The result combines tactile clay characters, a lava canyon, a large leap, and a forceful under-body camera pass.

English prompt translation

In clay-animation style, a running fox reaches a cliff and makes a committed heroic slow-motion leap over a vast lava canyon. During the jump, drive the camera rapidly beneath the body to reveal the depth of th...

Reference edit rhythm transfer MiniMax H3 result frame Actual H3 output / 4s
Multi-reference edit / Reference Generation

Reference edit rhythm transfer

Six supplied visuals are assembled into a compact edit that follows the reference clip rhythm and transitions.

English prompt translation

Use Images 1-6 as the content set. Follow Video 1 strictly for shot rhythm, transition style, cut order, and music timing. Preserve the supplied image identities while rebuilding the same editorial structure ar...

Selective voxel edit A MiniMax H3 result frame Actual H3 output / 4s
Video edit variant / Reference Generation

Selective voxel edit A

This short variant selectively transforms moving street objects while leaving people and architecture photographic.

English prompt translation

Keep the source buildings, pedestrians, and overall environment realistic. Convert only the trees and cars into 3D pixel or voxel blocks using Image 1 for style. Preserve their motion paths plus the real scene ...

Selective voxel edit B MiniMax H3 result frame Actual H3 output / 7s
Video edit variant / Reference Generation

Selective voxel edit B

The longer variant demonstrates the same selective voxel rule over a moving street shot.

English prompt translation

In Video 1, keep the buildings, pedestrians, pavement, camera motion, and natural environment live action. Change only trees and cars into Minecraft-like voxel forms based on Image 1, with correct trajectories,...

Hitchcock move and song transfer MiniMax H3 result frame Actual H3 output / 7s
Multi-video reference / Reference Generation

Hitchcock move and song transfer

The output combines a dramatic reference camera move with a new singer and transferred vocal performance.

English prompt translation

Use Video 1 only for the Hitchcock-style camera movement. Make the performer from Video 2 sing, while Video 3 defines the vocal performance and singing behavior. Keep the subject identity from Video 2 and align...

Three-character motion remap MiniMax H3 result frame Actual H3 output / 7s
Action replacement / Reference Generation

Three-character motion remap

Three replacement characters reproduce a fast choreographed floor routine and end in the same stacked formation.

English prompt translation

Use a fixed wide camera and replace the three suited performers in Video 1 with three realistic capybaras. Follow every original path exactly: all three drop, the left one jumps to center, the center rolls left...

Street-dance motion transfer MiniMax H3 result frame Actual H3 output / 10s
Action reference / Reference Generation

Street-dance motion transfer

The output transfers a fast street-dance routine to the character defined by the image references.

English prompt translation

Use Video 1 strictly for the street-dance movement and timing. Use Images 1 and 2 for the performer identity, appearance, and wardrobe. Recreate the complete dance performance with the referenced person while p...

Cloned-voice outdoor line MiniMax H3 result frame Actual H3 output / 10s
Voice transfer / Reference Generation

Cloned-voice outdoor line

The speaker delivers the new English line outdoors with the requested transferred voice color.

English prompt translation

Have the character say exactly: "Follow the wind, live free. Leave worries behind, enjoy the moment." Use Audio 1 only as the voice-color reference. Preserve natural outdoor performance, breathing, mouth timing...

Cat-to-dog replacement MiniMax H3 result frame Actual H3 output / 7s
Object edit / Reference Generation

Cat-to-dog replacement

The source interaction remains intact while the animal subject is replaced and integrated into the same camera move.

English prompt translation

In the source video, replace the cat with a dog. Preserve the person, camera path, approach, affectionate interaction, scene layout, timing, and all unedited visual content. Match the new animal scale, fur ligh...

Add a matching team member MiniMax H3 result frame Actual H3 output / 7s
Character edit / Reference Generation

Add a matching team member

A third uniformed person is added to the walking group and follows the same movement and environment.

English prompt translation

Add one person on the left side of the frame wearing the same team uniform. Match the movement, timing, lighting, scale, perspective, and slight weightless feeling of the people already in the source. Preserve ...

Character and wardrobe detail edit MiniMax H3 result frame Actual H3 output / 7s
Multi-object edit / Reference Generation

Character and wardrobe detail edit

The output performs both a character replacement and a targeted jacket change while retaining the group action.

English prompt translation

In Video 1, replace the child at the rear with the golden retriever from Image 1. Change the khaki jacket on the left child to the denim jacket from Image 2. Preserve every other person, action, camera movement...

Fairy-tale greenscreen composite B MiniMax H3 result frame Actual H3 output / 15s
Scene replacement variant / Reference Generation

Fairy-tale greenscreen composite B

A second official output applies the same greenscreen and relighting instruction to a distinct fantasy composite.

English prompt translation

Remove Video 1 green background and replace it with a fairy-tale environment like Video 2. Align the new environment with the performer action and relight the subject to match its color, direction, depth, and a...

Day-to-night relighting MiniMax H3 result frame Actual H3 output / 7s
Lighting edit / Reference Generation

Day-to-night relighting

The source action is retained while the scene becomes a coherent low-light nighttime environment.

English prompt translation

Use the reference video as the exact base and change its lighting to night. Preserve the person, action, environment geometry, camera, and timing. Rebuild exposure, color temperature, practical sources, shadows...

Window-view replacement MiniMax H3 result frame Actual H3 output / 7s
Location edit / Reference Generation

Window-view replacement

The interior shot stays intact while the visible exterior is replaced with a new referenced location.

English prompt translation

Replace only the view outside the window in Video 1 with the location shown in Image 1. Keep the interior, people, window frame, reflections, exposure, camera movement, and action unchanged. Match the new exter...

Dialogue and performance replacement MiniMax H3 result frame Actual H3 output / 10s
Audio edit / Reference Generation

Dialogue and performance replacement

The existing farewell scene is reinterpreted through a new spoken plea and a matching emotional performance.

English prompt translation

Replace the woman dialogue in Video 1 with the line from Audio 1: "Please do not leave. This time, let us not let go of each other." Adjust only the necessary facial and emotional performance to fit the new ple...

Multi-object precision edit MiniMax H3 result frame Actual H3 output / 10s
Instruction following / Reference Generation

Multi-object precision edit

The result executes multiple object, wardrobe, effect, and background edits in one continuous source sequence.

English prompt translation

In the reference video, replace the newspaper with a green-covered book and the chair with a red sofa. Remove the sunglasses and show the face clearly. Remove the burning-car effect so the vehicle is normal. Re...

Multi-scene beverage edit MiniMax H3 result frame Actual H3 output / 15s
Instruction following / Reference Generation

Multi-scene beverage edit

Product, signage, final bag contents, and the closing spoken line are changed consistently across several shots.

English prompt translation

Across the reference video, replace the opening canned drink with a red cola can, change the glowing convenience-store sign from its original name to "Huhui," replace every snack in the final plastic bag with m...

Magician suit swap MiniMax H3 result frame Actual H3 output / 7s
Precision performance / Text-to-Video

Magician suit swap

The result performs the synchronized smoke reveal, controlled suit-color exchange, bow, and curtain-color ending.

English prompt translation

Two magicians face the audience and perform a swap illusion. They wave their wands together and smoke rises. When it clears, their suit colors have exchanged: the left magician now wears white and the right mag...

Hand-drawn romance effects MiniMax H3 result frame Actual H3 output / 15s
Creative edit / Reference Generation

Hand-drawn romance effects

Animated sparks accumulate around a couple, brighten as they approach, and turn pink at the final intimate beat.

English prompt translation

Add orange-yellow hand-drawn marks around the two people in Video 1, using Image 1 for the doodle language. As they approach, increase the amount and intensity from small sparks to bright light. At the kiss, in...

Tram doodle transformation MiniMax H3 result frame Actual H3 output / 15s
Mixed-media story / Text-to-Video

Tram doodle transformation

A single glowing doodle races through a moving tram, repeatedly transforms, briefly remakes the carriage, and returns to paper.

English prompt translation

Create a 15-second handheld phone video inside an old tram at dusk. One warm apricot hand-drawn line continuously changes from a ticket into a paper swallow, caterpillar, arrow, sailboat, tiny tram, snail, umbr...

Fly Detectives trap MV A MiniMax H3 result frame Actual H3 output / 10s
Music video / Reference Generation

Fly Detectives trap MV A

The first variant delivers close rap performance, underground spaces, print-poster type, and beat-locked hard cuts.

English prompt translation

Create a 10-second 16:9 trap performance with two sharply styled detective partners. Use Image 3 for oppressive underground locations, Image 2 for high-impact print typography, and Image 1 for both performers. ...

Fly Detectives trap MV B MiniMax H3 result frame Actual H3 output / 10s
Music video variant / Reference Generation

Fly Detectives trap MV B

The second official variant preserves the same music-video system with alternate performance shots and typography timing.

English prompt translation

Use the same detective identities, underground scene reference, and typography system to create a second 10-second trap edit. Alternate face, hand, shoulder, and half-body coverage while phrases such as "TWO FL...

Community reference

More sourced H3 examples

Public creator posts can reveal useful prompt patterns, but they rarely disclose every input, parameter, retry, or edit.

Text-to-Video
Cliff-city speeder chase MiniMax H3 video cover 15 sec / X source
Umesh / @umesh_ai / Prompt in post

Cliff-city speeder chase

“Speeder chase across a cliff city (single continuous shot)”

A high-speed chase staged as one continuous camera move through a monumental city carved into cliffs.

Prompt pattern
Subject + environment + one continuous camera path. The prompt establishes scale before accelerating into the chase.
Community example / video served from X CDNView original post on X
Reference-to-Video
Five-second superhero intro MiniMax H3 video cover 22 sec / X source
Larus Canus / @MrLarus / Prompt in post

Five-second superhero intro

“Create a 5-second, 16:9, single-shot live-action superhero intro”

A short live-action character reveal built from a reference image, with the opening motion and final composition specified.

Prompt pattern
Duration + ratio + shot count + medium first, then reference ownership, opening motion, and final composition.
Community example / video served from X CDNView original post on X
Text-to-Video
Rainbow skunk supermarket leap MiniMax H3 video cover 5 sec / X source
Simon Willison / @simonw / Prompt in post

Rainbow skunk supermarket leap

““a rainbow colored skunk leaps over a mossy log in a supermarket””

A compact surreal action beat that tests whether a short prompt can preserve a colorful subject, trajectory, and setting.

Prompt pattern
One subject + one action + one unusual setting. The short prompt leaves H3 room to stage the movement.
Community example / video served from X CDNView original post on X
Local H3
Local 5090 test: space cat in rain MiniMax H3 video cover 15 sec / X source
新清士 / @kiyoshi_shin / Prompt in post

Local 5090 test: space cat in rain

“宇宙猫、雨の町を歩く / “Space cat walks through a rainy town””

A local 768P run on an RTX 5090, useful for setting expectations around generation time and a simple text-to-video prompt.

Prompt pattern
Minimal subject + setting + action. The post also records hardware, resolution, duration, and elapsed generation time.
Community example / video served from X CDNView original post on X
Poster / image-led
Poster motion study MiniMax H3 video cover 10 sec / X source
LudovicCreator / @LudovicCreator / Prompt in first comment

Poster motion study

“Here an example for poster. Find the image used and prompt in first comment.”

A portrait-format poster treatment made in MiniMax H3, showing how a still image can become a short motion concept.

Prompt pattern
Start with a designed still, then move the prompt and source image into a controlled poster reveal.
Community example / video served from X CDNView original post on X
Model comparison
Seedance 2 vs FLUX 3 vs H3 MiniMax H3 video cover 15 sec / X source
GENEL / @genel_ai / Prompt in replies

Seedance 2 vs FLUX 3 vs H3

“Seedance 2 vs FLUX 3 vs MiniMax H3”

A side-by-side comparison frame for studying how the same visual brief changes across video models.

Prompt pattern
Hold the subject and prompt constant, then compare motion, composition, and temporal stability across models.
Community example / video served from X CDNView original post on X
Image-to-Video
Kaiju attack from an airplane window MiniMax H3 video cover 10 sec / X source
Spectro / @Spectromachina / Prompt in post

Kaiju attack from an airplane window

“Handheld documentary-style footage, shot from inside a flying airplane window, slightly warped by the plexiglass...”

Handheld documentary footage through a warped airplane window, showing a colossal squid-like kaiju tearing through a city skyline until an explosion detonates against its flank.

Prompt pattern
Reference image + FL2VA with a clear POV and environmental detail. The prompt establishes the camera location before the action begins.
Community example / video served from X CDNView original post on X
Reference-to-Video
Interactive creature encyclopedia UI MiniMax H3 video cover 15 sec / X source
Kōda / @aimikoda / Prompt in post

Interactive creature encyclopedia UI

“Use Image 1 as the exact UI/layout/style reference for the creature encyclopedia screen. Use Images 2–9 as the exact creature references...”

A locked-camera, minimal creature encyclopedia UI where a cursor clicks cards, updates the title, and triggers subtle creature reactions ending with a comedic swallow.

Prompt pattern
Multi-image reference generation with strict timing, stable UI, no cuts, no camera move, and explicit audio cues for UI clicks and creature reactions.
Community example / video served from X CDNView original post on X
Text-to-Video
Kowloon Walled City one-take MiniMax H3 video cover 10 sec / X source
cocktail peanut / @cocktailpeanut / Prompt in post

Kowloon Walled City one-take

“Inside Kowloon Walled City, late 1980s, ultra-realistic handheld one-take. A restaurant worker hurries through impossibly narrow, crowded corridors...”

A claustrophobic, handheld one-take chase through 1980s Kowloon Walled City as a restaurant worker carries boiling soup through crowded corridors until the lights cut out.

Prompt pattern
Pure text-to-video with no start image and heavy negative guidance. Documentary-style realism cues control lighting, sound, and pacing.
Community example / video served from X CDNView original post on X
Text-to-Video
Luxury headphones product showcase MiniMax H3 video cover 15 sec / X source
LudovicCreator / @LudovicCreator / Prompt in post

Luxury headphones product showcase

“Create a 15-second luxury cinematic product showcase for premium wireless over-ear headphones...”

Cinematic macro tracking shot transitions to a hero rotation, an elegant exploded-view reveal of internal components, and a seamless reassembly with rim-lit final composition.

Prompt pattern
Time-segmented prompt specifying exact camera moves, materials, geometry constraints, and finishing look for product visualization.
Community example / video served from X CDNView original post on X
Use examples as lessons

Watch, isolate the control, then adapt one variable.

Start from the closest result, copy its structure, and replace the subject or reference roles first. Exact reproduction may still require the original references, settings, seed, and attempt history.

Build a controlled prompt