How to Replace a Character in Video with MiniMax H3 Without Rebuilding the Shot

Quick Answer

To replace a character in a video with MiniMax H3, use the original video as the performance reference and the new character image as the identity reference.

The key is to separate two jobs:

• The video controls what happens: movement, timing, camera, and scene.
• The character reference controls who appears: face, hairstyle, outfit, and identity.

In my test, replacing the character from an existing video preserved the original close-up shot better than creating a new video with the same setup. The final result kept the original performance while changing Carrie into a new character.

What You Need Before You Start

A source video with the performance you want to keep

Character replacement works best when you already have a video that contains the motion and camera behavior you want.

For this workflow, I used an existing MiniMax H3 base video.

The clip showed Carrie in an indoor living room scene. It used a close camera angle focused on her face and upper body. The shot included a small head movement and a subtle turn toward the camera.

This type of clip is useful because identity changes are easier to evaluate when the face is clearly visible.

A character reference for the new identity

Before replacing the character, prepare the new character assets.

For this test, I created a new character and saved the main image as:

@hero

I also prepared additional views:

• @CHAR2_Front
• @CHAR2_34
• @CHAR2_Side

The first test only used the main character image.

The goal was to see whether one clear identity reference was enough before adding more inputs.

A clear replacement goal

Before generating, define what should change and what should stay.

In this workflow:

Change:

• Carrie's identity
• facial features
• hairstyle
• character appearance

Keep:

• camera framing
• movement
• timing
• room environment
• lighting
• overall shot structure

This prevents the generation from becoming a completely new video.

How to Replace a Character in Video with MiniMax H3

Step 1: Start with an approved base video

The first step is not creating a new clip.

Instead, start with a video that already works.

For this project, I reused my existing VID_01_BASE clip instead of generating another scene.

The reason was simple: the original video already had the motion, composition, and camera behavior I wanted. Rebuilding the shot would introduce unnecessary changes.

Step 2: Add the new character reference

Next, connect the new character reference to the replacement workflow.

For the first attempt, I used:

@hero

as the identity reference.

The reference image was responsible for defining the new character.

The video remained responsible for the original performance.

This separation is the most important part of the workflow.

Step 3: Use the source video as the performance reference

The prompt should clearly explain the relationship between the video and the new character.

My prompt focused on three instructions:

1. Replace Carrie with the new character.
2. Keep the original video structure.
3. Preserve the original movement and environment.

The final prompt direction was:

Use @hero as the new character reference.

Replace Carrie in the source video with @hero.

Keep the original shot structure, close framing, face and upper-body composition, body movement, timing, camera stability, indoor living room background, lighting, and overall visual continuity unchanged.

Preserve the subtle turn toward the camera and the natural expression from the original clip.

The final result should look like the same shot and the same performance, but with @hero replacing Carrie.

The important part is not adding more description about the new character.

The important part is telling the model what information comes from the video and what information comes from the character reference.

Step 4: Generate the replacement video

After connecting the source video and character reference, generate the new version.

In my test, the replacement worked as expected.

The new character appeared in the same close-up setup. The face was clearly different from Carrie, while the original shot structure remained recognizable.

The most noticeable moment was the transition from side view to front view.

At the beginning of the clip, the new character appeared from a side angle with the hair moved behind the ear. When the character turned toward the camera, the new identity was still clearly visible.

This made it easier to confirm that the replacement was not just a surface change.

Step 5: Check the result with three QA points

A successful replacement is not only about whether the face changed.

I checked the result in three areas.

Identity consistency

First, check whether the new character remains recognizable.

Look at:

• facial features
• hairstyle
• overall appearance

In my result, the character identity stayed consistent between the side view and the front-facing view.

Performance preservation

Next, compare the movement.

The purpose of using the original video is to keep the existing performance.

Check:

• head movement
• timing
• expression changes
• camera behavior

The replacement result kept the original close-up performance instead of creating a completely new action.

Scene continuity

Finally, check the environment.

A good replacement should not accidentally rebuild the entire scene.

Check:

• background layout
• lighting
• framing
• camera distance

The successful result kept the indoor living room environment and the original shot style.

Why Character Replacement Worked Better Than Regenerating the Shot

Regenerating a similar video is not the same as replacing a character

Before using the replacement workflow, I tried another approach.

I kept the original prompt structure and changed the character references to the new character.

The result created a new video with the new character.

However, this was not the same as replacing a person inside an existing video.

The generation understood the scene concept, but it recreated the shot instead of preserving the original performance.

This approach can work when you want a new video with a similar style.

It is less suitable when you need the exact movement and camera continuity from an existing clip.

A source video gives the model a stronger performance reference

The successful workflow used:

Existing video
+
New character reference

instead of:

New character reference
+
Text description of the scene

The difference was visible.

The replacement workflow kept the original shot while changing only the character identity.

For tasks like character swaps, this creates a clearer production process:

• build the performance once
• reuse it
• change the character when needed

Close-up shots make replacement easier to evaluate

The base video in this test used a close camera angle.

This turned out to be useful because the character identity was easy to check.

A wider shot may hide small identity changes. A close-up reveals whether the replacement remains consistent during movement.

For character replacement tests, a simple shot with a clear face is often more useful than a complicated action scene.

Common Problems and Fixes

The original character is still visible

If the output still looks like the original person, the workflow may not be treating the task as a replacement.

Try:

• using the video as the source reference
• adding a clear character identity reference
• explicitly saying that the original character should be replaced

The result looks like a new video instead of a replacement

This usually happens when the workflow focuses too much on describing the scene.

The model may recreate:

• the room
• the camera
• the action

instead of preserving the original clip.

Use the source video as the main reference and describe what should remain unchanged.

The character changes during movement

A single reference image may not provide enough information for every angle.

If the character becomes unstable:

• add a front view
• add a three-quarter view
• add a side view when the shot requires it

For example:

@hero
+
@CHAR2_34

can provide more information about the character from different angles.

The video changes too much

If the background or framing changes, reduce unnecessary instructions about new content.

The prompt should focus on preservation:

• keep the original shot
• keep the original camera
• keep the original environment

Character replacement works best when the model has fewer reasons to redesign the scene.

FAQ

Should I regenerate a video or replace the character in an existing video?

Choose regeneration when you only need a similar scene with a different character.

Choose replacement when you want to keep the original movement, camera, and timing.

The two workflows solve different problems.

Can I replace a character with only one reference image?

For simple shots, one clear character image may be enough.

In my test, a single main character reference was enough because the source video was a close-up shot.

For more complex movement or wider shots, additional views may help.

What type of video works best for character replacement?

A useful source clip usually has:

• a visible character
• clear movement
• stable camera behavior
• a simple environment

Close-up and medium shots are easier to evaluate because identity changes are more obvious.

Is video editing the same as character replacement?

Not always.

In my test, using a general video edit workflow did not replace the character successfully.

The replacement workflow worked better because it treated the original video as the performance source and the new character as the identity source.

Can I reuse the same source video for different characters?

Yes, this is one of the biggest advantages of the workflow.

Once a useful performance clip exists, you can reuse it as a base and create different character versions without rebuilding the entire shot.