MiniMax H3 Motion Transfer: Complete Guide

MiniMax H3 Motion Transfer lets creators use the movement, timing, or performance from an existing video as guidance for a newly generated clip. Instead of trying to describe a complicated dance, fight sequence, body gesture, or performance frame by frame, you can provide a reference video and tell MiniMax H3 which character should perform that motion.

The simplest version of the workflow is:

Reference image = who performs the action

Reference video = how the character moves

That makes MiniMax H3 Motion Transfer useful for character animation, dance videos, game characters, virtual influencers, action scenes, fashion clips, stylized avatars, and other situations where the desired movement is much easier to show than to describe.

MiniMax officially highlights V2V motion transfer as one of H3’s strengths. H3 can understand text, images, video, and audio in a shared multimodal context, while its Reference mode can use references for characters, motion, camera behavior, style, voice, and editing rhythm.

This guide focuses specifically on how MiniMax H3 Motion Transfer works, how to choose a good motion reference, how to separate character motion from camera motion, how to write better prompts, and how to troubleshoot results when the transferred movement does not behave as expected.

What Is MiniMax H3 Motion Transfer?

MiniMax H3 Motion Transfer is a reference-based video workflow in which an existing video provides motion information for a new generation.

Imagine that you have:

• a still image of a futuristic robot;

• a reference video of a person performing a martial-arts sequence;

• a prompt describing a new environment.

Instead of asking the model to reproduce the person in the reference video, the goal is to keep the robot’s appearance while borrowing the movement and timing from the performer.

A basic prompt might say:

Use Image 1 as the character reference. Preserve the robot’s armor design, proportions, helmet, and illuminated blue details. Follow the body movement, turns, timing, and combat performance from Video 1. Place the character on a futuristic rooftop at night. Keep the robot’s appearance consistent throughout the shot.

This distinction matters.

The reference image answers:

Who or what should appear?

The reference video answers:

How should that subject move?

The text prompt explains how those two sources should be combined.

MiniMax’s H3 launch materials describe this more broadly as generalized reference and editing, where relationships between references and the target video can be expressed through natural language rather than being limited to one fixed task.

How MiniMax H3 Motion Transfer Works

A useful way to understand the workflow is to separate it into three layers, each owned by a different input.

The result is not simply “copy this video.”

The stronger instruction is:

Keep this subject, borrow this motion, and create a new scene around them.

MiniMax H3 Motion Transfer vs Reference to Video

MiniMax H3 Motion Transfer and MiniMax H3 Reference to Video are closely related, but they do not represent exactly the same search intent.

Reference to Video is the broader workflow.

It may use:

• character references

• product references

• style references

• motion references

• camera references

• audio references

Motion Transfer is a narrower task inside that workflow.

The user’s goal is specifically:

“Take the movement from this video and apply it to another subject.”

This is why a dedicated MiniMax H3 Motion Transfer workflow is useful even if you already understand Reference to Video.

If the reference video is there mainly because of what the person does, you are working with motion transfer.

If the reference video is mainly there because of how the camera moves, you are working with camera reference.

If both matter, the prompt should say so explicitly.

How to Use MiniMax H3 Motion Transfer

Step 1: Choose the Target Character

Start with a clear image of the subject that will perform the transferred motion.

A good character reference should make important identity details easy to see.

For human characters, a useful image usually shows:

• a clear face;

• recognizable clothing;

• useful body proportions;

• minimal obstruction.

For robots, creatures, or fictional characters, prioritize:

• readable silhouette;

• limbs that can plausibly perform the action;

• clear costume or armor details;

• consistent proportions.

Motion becomes harder to transfer cleanly when the target subject has a body structure that is radically different from the performer.

Step 2: Choose a Clean Motion Reference

The reference video is one of the most important parts of MiniMax H3 Motion Transfer.

Look for a clip where the intended movement is easy to understand.

A useful motion reference often has:

• one main performer;

• visible arms and legs;

• limited occlusion;

• a continuous action;

• relatively clear timing;

• few or no cuts.

For example, if the desired output is a robot performing a dance, a clean full-body dance video will usually communicate the motion better than a heavily edited music video containing six camera cuts.

The goal is to reduce ambiguity.

Step 3: Decide Whether You Want the Camera Motion Too

This is one of the most important decisions in MiniMax H3 Motion Transfer.

Subject motion and camera motion are different things.

Suppose Video 1 contains a dancer moving forward while the camera rapidly orbits around them.

You may want:

A. The dancer’s movement only

or:

B. The movement plus the orbiting camera

These require different prompts.

For motion only:

Follow the performer’s body movement and timing from Video 1. Do not copy the camera movement. Use a stable medium full-body shot.

For both:

Follow the performer’s body movement and timing from Video 1, including the orbiting camera movement and changing viewing angle.

MiniMax’s official H3 example demonstrates that camera movement itself can be explicitly assigned to a reference video, reinforcing why camera and subject motion should be described separately.

Step 4: Assign the References Clearly

Avoid a vague instruction such as:

Use Image 1 and Video 1 to make a cool action scene.

A stronger MiniMax H3 Motion Transfer prompt says:

Image 1 defines the character. Video 1 defines the body movement and timing. Preserve the character identity from Image 1 throughout the entire sequence.

This tells the model which information should come from each asset. In the generator, you refer to uploaded assets in the prompt as @Image, @Video, and @Audio.

Step 5: Add the New Scene After the Transfer Instruction

Once identity and motion roles are clear, add the creative direction.

For example:

Use Image 1 as the character reference. Follow the dance movement and timing from Video 1. Preserve the character’s face, hairstyle, silver jacket, and proportions. Place her inside a neon-lit subway station at night. Use realistic fabric motion and cinematic reflections. Keep the camera in a smooth medium full-body tracking shot.

The hierarchy is:

character → motion → consistency → scene → camera → style

That order tends to keep the core task clear.

Try MiniMax H3 Motion Transfer with your own character image and reference video.

MiniMax H3 Motion Transfer Prompt Formula

A reusable formula is:

Target subject + identity constraints + motion source + motion details + camera behavior + environment + style + consistency constraints

Template:

Use Image 1 as the target character reference. Preserve [face / outfit / body proportions / design details]. Follow the [body movement / gesture / dance / performance / timing] from Video 1. [Copy / do not copy] the camera movement from Video 1. Place the character in [environment]. Use [lighting / visual style]. Keep [important identity details] consistent throughout the entire shot.

The most important sentence is often:

Follow the body movement and timing from Video 1 while preserving the character identity from Image 1.

That communicates the central MiniMax H3 Motion Transfer relationship.

MiniMax H3 Motion Transfer Prompt Examples

Example 1: Dance Motion Transfer

Goal: Transfer a human dance performance onto a new character.

Use Image 1 as the character reference. Preserve her facial appearance, long black hair, red jacket, black trousers, and body proportions. Follow the full-body dance movement, footwork, arm gestures, turns, and performance timing from Video 1. Do not copy the original background. Place her inside a modern neon studio with a reflective floor. Use a stable full-body camera that keeps the complete dance visible. Maintain identity and clothing consistency throughout.

This structure works well when the motion itself is the priority.

Notice that the prompt explicitly rejects the original background.

That helps distinguish:

what to borrow from what to regenerate.

Example 2: Robot Combat Motion Transfer

Goal: Apply a martial-arts performance to a robot.

Use Image 1 as the robot character reference. Preserve the armor silhouette, helmet shape, metallic panels, blue illuminated details, and mechanical proportions. Follow the combat movements, punches, turns, defensive gestures, and timing from Video 1. Place the robot on a rain-soaked futuristic rooftop at night. Use dramatic side lighting and realistic mechanical articulation. Keep the robot design stable across every pose.

This type of MiniMax H3 Motion Transfer is useful when a fictional character needs movement that would be difficult to describe precisely through text alone.

Example 3: Fashion Walk Motion Transfer

Goal: Transfer a model’s runway movement to a different character.

Use Image 1 as the fashion character reference. Preserve her face, hairstyle, long red dress, silhouette, and fabric design. Follow the walking rhythm, posture, arm movement, and turns from Video 1. Place her inside a minimalist concrete gallery. Use soft editorial lighting and a smooth backward tracking camera. Keep the dress design and facial identity consistent while allowing natural fabric movement.

This workflow separates:

• character identity;

• runway performance;

• new environment;

• new camera direction.

Example 4: Motion Plus Camera Transfer

Sometimes you want both the performance and the original cinematography.

Prompt:

Use Image 1 as the new character reference. Preserve the character’s identity, clothing, and proportions. Follow both the body performance and the circular camera movement from Video 1. Match the timing of the subject turns with the camera orbit. Replace the original location with a futuristic underground station illuminated by red and white practical lights. Keep the new character recognizable from every viewing angle.

Here the prompt explicitly asks for two different things from Video 1:

1. subject performance;

2. camera path.

This is much clearer than simply saying “copy Video 1.”

How to Improve MiniMax H3 Motion Transfer Results

Use a Motion Reference With One Clear Action

If a short reference video contains:

• running;

• sitting;

• jumping;

• dancing;

• multiple cuts;

the requested motion becomes harder to interpret.

A focused motion sequence is easier to assign.

For example:

one continuous spin + landing

is clearer than:

an entire edited action montage.

Keep the Target Character Visible

Motion transfer becomes more difficult when the identity reference provides only a close-up face while the motion reference shows complex full-body movement.

If body mechanics matter, give the model useful body information.

Tell H3 What Not to Copy

A reference video also carries its original performer, background, clothing, lighting, camera behavior, and style.

If only the movement matters, say so.

For example:

Use only the body movement and timing from Video 1. Do not copy its character appearance, clothing, background, or camera motion.

H3 has no separate negative-prompt field, so exclusions like this belong in the prompt as ordinary sentences.

Describe Critical Motion, Not Every Frame

The reference already communicates temporal detail. The prompt only needs to name the parts that matter — a fast spin, a natural walking rhythm, sharp combat gestures, a slow expressive hand motion, a synchronized jump.

Do not try to rewrite the entire reference video frame by frame unless a specific detail needs emphasis.

Keep Identity Constraints Short and Specific

Instead of:

Make sure the character looks exactly the same all the time and everything about them stays consistent.

Write:

Preserve the face, hairstyle, red jacket, body proportions, and black boots throughout.

Specific constraints are easier to interpret.

Common MiniMax H3 Motion Transfer Problems

The general rule is:

If MiniMax H3 Motion Transfer fails, first identify whether the failure is in identity, body movement, camera movement, or scene control.

Do not rewrite everything at once.

Motion Transfer vs Camera Motion Transfer

This difference deserves its own section because many reference videos contain both.

Motion Transfer

The target is the subject’s physical performance — what the body does.

Follow the subject’s body movement from Video 1.

Camera Motion Transfer

The target is cinematography — how the camera behaves, independent of the subject.

Reference the camera movement from Video 1.

MiniMax’s own H3 demonstration uses natural-language instructions to reference a specific camera movement from one video while assigning a character from an image and vocals from audio, showing that different multimodal references can be given separate roles.

Need Both?

Follow the body performance and camera movement from Video 1.

The clearer the assignment, the easier it is to understand what should survive in the output.

MiniMax H3 Motion Transfer Input Limits

MiniMax H3 treats motion references as part of its broader Reference mode. In the generator, a single job accepts up to 9 reference images, up to 3 reference videos (MP4/MOV), and up to 3 reference audio clips, for a maximum of 12 reference files in total. Reference audio requires an accompanying image or video.

Output settings are selected in the generator before you submit, and the interface shows the credit cost for the settings you pick. Duration and aspect ratio options appear in the controls under the prompt box, so check the current selections there rather than assuming a fixed value.

These limits make it possible to build complex motion-transfer setups, but complexity should be added only when it serves a clear purpose.

For most tests, one character image and one strong motion reference are enough.

When Should You Use MiniMax H3 Motion Transfer?

MiniMax H3 Motion Transfer is a good choice when your idea sounds like:

• “Make this robot perform that dance.”

• “Put this character into that fight choreography.”

• “Use this actor’s gestures for my fantasy character.”

• “Use the body movement from this video without copying the original person.”

It is less necessary when the motion can already be described very simply.

For example:

“The woman slowly turns her head.”

A text prompt may be enough. Motion transfer becomes more useful for choreography, coordinated full-body movement, distinctive gestures, subtle performance, and timing that is difficult to describe precisely.

If your goal is holding one character recognizable across several different shots rather than borrowing a movement, see our guide to keeping a consistent character in MiniMax H3. For an overview of how Reference mode sits alongside the other generation modes, see how to use MiniMax H3.

FAQ

What is MiniMax H3 Motion Transfer?

MiniMax H3 Motion Transfer uses a reference video to guide the movement or performance of a newly generated subject. A separate image can define the target character while the reference video provides body motion and timing.

Can MiniMax H3 copy motion from a video?

MiniMax officially lists V2V motion transfer and motion reference among H3’s capabilities. Reference mode can use video as motion context while a prompt explains how that movement should be applied to the target generation.

How do I transfer motion to another character in MiniMax H3?

Use a clear image for the target character and a reference video containing the desired movement. In the prompt, explicitly state that the image defines character identity while the video defines body movement and timing.

Can MiniMax H3 preserve the character while transferring motion?

Reference images can be used to guide character identity while video references guide motion. To improve consistency, explicitly list the character features that should remain unchanged and avoid giving conflicting identity instructions.

Why is MiniMax H3 not following my motion reference?

Common causes include an unclear reference assignment, conflicting text instructions, a heavily edited reference video, or a motion sequence that is difficult to map onto the target character. Simplify the prompt and state clearly that the reference video defines motion and timing.

Does MiniMax H3 copy camera movement too?

A video reference can provide camera information as well as subject motion. If you want only body movement, explicitly tell H3 not to copy the camera motion. If you want both, specify both roles.

How many motion reference videos can MiniMax H3 use?

Reference mode accepts up to three reference videos in a single job, alongside up to nine images and three audio clips, with a maximum of twelve reference files in total.

What is the best MiniMax H3 Motion Transfer prompt structure?

A useful starting structure is:

target character + identity details to preserve + reference video motion + whether to copy camera movement + new environment + visual style + consistency constraints

The most important step is clearly separating who the subject is from how the subject moves.

Final Thoughts

MiniMax H3 Motion Transfer is most useful when a movement is easier to demonstrate than to describe.

Instead of trying to explain every step of a dance, fight sequence, gesture, runway walk, or character performance in text, a reference video can supply the temporal information directly.

The strongest workflow keeps the roles simple:

Image 1 defines the character.

Video 1 defines the motion.

The prompt explains what to preserve, what to replace, and how the new scene should look.

For the best results, start with one clear character reference and one clean motion reference. Separate body movement from camera movement, state the identity details that cannot change, and avoid adding unnecessary references until the basic transfer works.

That approach makes MiniMax H3 Motion Transfer especially useful for game characters, virtual influencers, fashion videos, stylized avatars, action scenes, dance clips, and other reference-driven video workflows where precise movement matters more than a generic text-generated action.

Create your own MiniMax H3 Motion Transfer video with a character image and motion reference.