MiniMax H3 is a multimodal video model from MiniMax. Animate an opening image, connect first and last frames, or combine image, video and audio references. On WeShop, the standard H3 selections offer 5–15-second requests.
Explore a city dance sequence, a game-style character customization scene, and a fashion campaign assembled from four references. Compare the supplied images, full original prompts and generated videos.
One Character across Six City Shots
The prompt specifies a yellow dress, ponytail and six city locations. The supplied portrait has a different hairstyle and outfit; the video follows the requested styling across changing landmarks. Compare appearance and shot timing with the request.
subject_definitions:
<Subject 1> is the young woman in <Picture 1>,
with a dark ponytail and a yellow sundress.
integrated_multimodal_description:
[Shot 1] <Subject 1> dances with Rome in the background.
[Shot 2] At 00:01.000, the shot changes to Paris.
[Shot 3] At 00:02.000, the shot changes to London.
[Shot 4] At 00:03.000, the shot changes to New York.
[Shot 5] At 00:04.000, the shot changes to Tokyo.
[Shot 6] At 00:05.000, the shot changes to Berlin.
overall_soundscape:
Light city ambience remains underneath the dance performance.
non_diegetic_music:
An upbeat instrumental dance rhythm without vocals.
Two images supply a stylized character and a game-menu design. The timed prompt directs equipment changes, interface panels and a transition into a cyberpunk street. Watch the sequence and check the generated interface text.
Character reference image 1, UI style reference image 2. [0 seconds - 2 seconds] High-angle overhead shot. The character sits on a highly saturated bright purple ground, as shown in image 1, looking up at the camera. The game menu UI is displayed on the right: Start New Game, Continue Game (highlighted), Settings, Exit Game. The player profile MINIMAX is displayed in the upper left corner. Click "Continue Game". [2 seconds - 4 seconds] Smoothly zoom to her right arm. The UI panel slides in from the right, and the "Right Arm EQUIPMENT" panel appears. Select the highlighted "Phantom Grap", then slide to "Chronos Claw". Her right hand mechanically reconfigures, fingers spread, new claw-like knuckles lock in place, and the cyan LED lights flash brighter. [4 seconds - 7 seconds] The camera smoothly moves to her left side. The new UI slides into the "ARMAMENT CUSTOMIZATION" grid, displaying the hand, forearm, elbow, and upper arm components. Selections quickly cycle through the components. Her left arm is disassembled piece by piece: the forearm plate detaches, new armor slides in, the elbow joint is replaced, and the hand is reconfigured, revealing exposed wiring and pistons during the replacement process. [7-8.5 seconds] The camera pulls back to a medium shot. The CONFIRM CONFIG button flashes. Click it. All UI panels shrink inward and disappear. The character's legs loosen, she adjusts her posture, and now sits relaxed, one knee raised, her prosthetic hand naturally raised, displaying a slight post-configuration movement. [8.5-10 seconds] A loading bar appears at the bottom. It quickly fills from 0% to 100%. The highly saturated bright purple environment begins to darken, shadows spread from the edges, and a warm golden light seeps in. [10-15 seconds] As she stands up, the complete environment loads around her. A dense cyberpunk slum appears, neon lights flashing, damp streets reflecting light, crowds surging, motorcycles weaving, and a tangled web of wires. Stacked buildings stretch towards futuristic skyscrapers in the distance. The camera stabilizes in a third-person perspective, positioned behind her, showing her back. HUD elements fade in, displaying a minimap in the upper right corner and a health and ammo counter in the lower left. A mission marker appears. She steps into the street.
Four images separately supply the office, person, suit and shoes. The result combines a black-suit portrait with an office setting and a shoe close-up; inspect the garment and product details as the shot changes.
The model in Figure 2 is wearing the suit in Figure 3 and the shoes in Figure 4, shooting an advertising campaign in the scene in Figure 1.
MiniMax H3 combines text, images, video and audio context for video creation and transformation. MiniMax describes motion transfer, multi-shot generation and native stereo sound. Use it for animated posters, product demonstrations, fashion scenes and game-style sequences; this page offers frame-led and reference workflows.
References with Distinct Roles
Combine a person, outfit, product or scene by describing what each input contributes. The four-reference fashion example brings an office, person, suit and shoes into a campaign video.
Directed Multi-Shot Sequences
Describe actions and scene changes in order. The city sequence changes landmarks, while the character-and-interface example moves from customization to a street scene. Check continuity and timing through the whole clip.
Native Stereo Sound
MiniMax H3 supports sound generated alongside video. Describe the desired ambience, music or effects, as in the city and poster prompts, then listen to the finished track and check its timing.
WORKFLOWS
Choose Your Starting Input
Choose MiniMax H3 Image to Video for an opening image and optional ending image. Choose MiniMax H3 Reference for a text brief with optional image, video or audio references. Switch between these two standard H3 workflows in the Model menu.
Animate an Opening Image
Upload the frame that establishes the subject and scene. Describe the action, environmental movement, or camera direction that follows it.
Guide the Opening and Ending Frames
Supply both images, then describe the action or transition connecting their compositions. Choose a duration offered for this mode.
Direct a Scene with References
Select MiniMax H3 Reference and add the media that should guide the video. Describe which input supplies appearance, motion, setting or sound. Choose Adaptive or a listed aspect ratio, then set the duration. Video references use Adaptive framing.
PROMPT GUIDE
How to Direct MiniMax H3 with a Prompt
Read each complete original prompt with its paired inputs and result. These examples show how to describe design boundaries, reference roles, subject motion and camera movement. Adapt the instructions to the materials you upload.
Specify the Design Elements to Keep
The prompt names the frame, layout and figure while requesting animation and sound. Compare the pale-green reference with the red result: the layout remains recognizable, but the palette changes. Check colors and lettering when adapting a poster.
Create a dynamic poster video that retains the original image's gallery white frame, inner frame, red, white and black color scheme, 3D figure feel and layout structure, and adds a lively sound effect when the text appears.
Assign Product Details and Appearance
The first reference is a product-information sheet and the second is a chair image. The prompt separates their roles, then requests detail views, adjustment demonstrations and a closing line. The clip includes a dark chair, airflow graphics and a spine diagram; review product fidelity and text.
360-degree display of Herman Miller black ergonomic chair for high-end offices. Breathable mesh back for close-up visualization of air flow. Lumbar support + ergonomic curve engineering animation. Multi-dimensional armrest/seat height adjustment demonstration. Designers/programmers/creatives work in scenarios where they work for a long time, and their posture can be relaxed and comfortable. 3D skeletal support diagram to prevent fatigue from sitting for long periods of time. Spatial aesthetics interior display. At the end, the words "Inspiration and comfort are online at the same time" appear. High-end simplicity, cool colors, professional futuristic feel, and slow-paced camera movement. Please refer to Figure 1 for product details and Figure 2 for product images.
Separate Subject Motion from Camera Motion
The crane turns its head while the prompt asks the camera to push from a wide shot toward a medium view. It also describes mist, ripples and natural sound without music. State these instructions separately, then check framing, animal motion and the soundtrack.
15 seconds, 16:9, realistic nature-documentary style.
In a misty wetland at dawn, an adult white crane stands on one leg in shallow water
and slowly turns its head toward the camera.
Soft backlight passes through the mist, and small ripples move across the water.
The camera slowly pushes from an extreme wide shot to a medium shot without cutting.
Use only wind, water, and distant bird calls. Do not add background music.
Describe the Model’s Forward Movement
The reference establishes a white runway outfit. The prompt requests a natural forward walk and clear clothing, with a constraint on the skirt lining. Watch the garment throughout the walk, including the moving hem.
Generate a smooth animation of the model's runway show, moving forward naturally, showing the clothes clearly. The movie has extremely high-quality visuals. There is an inner lining in the skirt hem, and it must not be visible.
HOW TO USE
Create Your First MiniMax H3 Video
Use the generator at the top of this page. Choose the input mode, prepare your brief and review the request before generating.
Provide the Input for This Model
Choose Image to Video and upload an opening image, adding an ending frame if needed. For a reference-led or text-led brief, choose MiniMax H3 Reference and describe the role of each supplied image, video or audio file.
Check the Available Settings and Points
Choose 5–15 seconds and the result quantity. Reference mode also offers Adaptive, 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. Review the points estimate and your account’s access before selecting Generate. A video reference uses Adaptive framing; check the form again after attaching media.
Open and Check the Result
Watch the whole clip for changes to faces, clothing, products, shot order and the ending composition. Listen to the soundtrack and check its timing. Download the result to inspect its actual dimensions and duration.
Choose standard MiniMax H3 for a 5–15-second scene. MiniMax H3 Lite provides a separate five-second workflow. Both have frame-led and reference selections; open the version you want, then choose its input mode.
Dimension
MiniMax H3
MiniMax H3 Lite
Model role
Standard H3 video creation and transformation
Separate H3 Lite selection for five-second requests
WeShop inputs
Opening image; optional ending image; or text with optional image, video and audio references
Opening image; optional ending image; or text with optional image, video and audio references
Current WeShop options
5–15 seconds; Reference mode aspect-ratio options
5 seconds; Reference mode aspect-ratio options
Model role
MiniMax H3
Standard H3 video creation and transformation
MiniMax H3 Lite
Separate H3 Lite selection for five-second requests
WeShop inputs
MiniMax H3
Opening image; optional ending image; or text with optional image, video and audio references
MiniMax H3 Lite
Opening image; optional ending image; or text with optional image, video and audio references
Current WeShop options
MiniMax H3
5–15 seconds; Reference mode aspect-ratio options
MiniMax H3 Lite
5 seconds; Reference mode aspect-ratio options
Choose MiniMax H3 When
Your scene needs a duration from 5 to 15 seconds, an opening and optional ending frame, or references with different creative roles. Use the standard H3 form at the top of this page.
You want the separate Lite selection for a five-second scene. Open the Lite page, then choose its image-to-video or reference mode and prepare the inputs for that mode.
Answers about this version, its inputs, and how to get started.
What is MiniMax H3?
MiniMax H3 is MiniMax’s multimodal video model for creation and transformation using text, images, video and audio context. On this page, choose standard H3 Image to Video or Reference.
What is MiniMax H3 useful for?
Use it for animated posters, product demonstrations, fashion scenes and game-style sequences. The examples show a city dance request, interface-driven character customization and a four-reference fashion campaign.
Which inputs can I use on this page?
Image to Video uses an opening image and optionally an ending image. Reference accepts a text brief with optional image, video and audio references. Explain what each reference should contribute to the new video.
Which durations, aspect ratios and resolutions can I choose?
Standard H3 offers 5–15 seconds. Reference mode also lists Adaptive, 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. There is no separate resolution selector in this form. Check the downloaded file for its actual pixel dimensions and duration. With a video reference, framing is Adaptive.
How should I write a MiniMax H3 prompt?
Identify the subject and the role of each reference, then describe the action, setting, camera movement and intended sound. For several shots, state their order. The fashion example assigns four images to the scene, person, suit and shoes; the crane prompt separates head movement from a camera push.
How do I choose between H3 and H3 Lite?
Use standard H3 for a 5–15-second request. Use the separate H3 Lite page for a five-second request. Each has image-to-video and reference selections; keep the mode and uploaded materials matched to your chosen version.
How do I check the cost?
Review the points estimate for the selected duration and quantity, then check WeShop Pricing and your account for model access and available credits before generating. The quote can change with the request.
Can it preserve a person, product or poster exactly?
References guide a newly generated video. Faces, clothing, product geometry, colors and text can change. Compare the full result with the supplied materials, especially through motion and shot transitions.
Can I generate without uploading an image?
Choose MiniMax H3 Reference for a text-led request; reference uploads are optional in that mode. The crane example uses a text prompt without reference images. Image to Video requires an opening image.
Can I direct the sound?
Describe the desired speech, ambience, effects or music in the prompt. H3 supports native stereo audio-video generation; listen to the output and check the words, timing and sound balance before using it.