MiniMax H3 Ref2Video Guide: How to Keep Characters Consistent Across Shots

Quick Answer

Keeping a character consistent across AI video shots is not about adding as many reference images as possible.

A better workflow is to build a stable character reference system first, then reuse the same identity anchors across different shots. Add extra references only when a new shot requires visual information that the current references do not provide.

In this workflow, I used the same character assets across three shots:

Stable identity references → generate a shot → check consistency → adjust only when needed → continue building new shots

The key is to separate what should stay fixed from what can change.

The character identity should remain consistent:

• facial features
• hairstyle
• outfit
• body proportions

The shot can change:

• camera angle
• framing
• movement
• expression

What You Need Before Creating Consistent Character Videos

Before generating multiple shots, prepare reusable character and scene references.

For this workflow, I created a reusable character reference system for Carrie.

The character assets included:

• @CHAR_01_HERO for facial identity
• @CHAR_FULL for body proportions and outfit
• @CHAR_SIDE for side-angle information

I also prepared multiple views of the same indoor environment:

• @SCENE_01_MED
• @SCENE_01_SIDE
• @SCENE_01_CLOSE

The purpose of these references is different.

A character reference answers:

Who is this character?

A scene reference answers:

Where is this shot happening?

Keeping these two parts separate makes it easier to find the cause when something changes.

For example, if the face changes, the issue may come from character references.

If the room changes, the issue may come from scene consistency.

How to Keep Characters Consistent Across MiniMax H3 Ref2Video Shots

Step 1: Create One Approved Character Baseline

The first step is not creating multiple variations.

Start with one reliable shot that establishes the character.

For my first shot, I used:

@CHAR_01_HERO
@CHAR_FULL
@SCENE_01_MED

The goal was simple:

Create a clear version of Carrie that could become the reference point for later shots.

The prompt focused on:

• natural movement
• stable appearance
• consistent outfit
• consistent environment

After generating the clip, I checked four things:

• Does the face still look like Carrie?
• Is the hairstyle unchanged?
• Is the outfit consistent?
• Does the body shape feel correct?

This shot became the identity baseline for the rest of the workflow.

Step 2: Test Whether an Additional Reference Actually Helps

A common assumption is that adding more reference images will always improve consistency.

I tested this during a side-angle shot.

The goal was not to create a better-looking video. The goal was to answer one production question:

Does a side-profile reference solve a real consistency problem?

I created two versions.

Version A: Use the Stable Identity References

The first version used:

@CHAR_01_HERO
@CHAR_FULL
@SCENE_01_SIDE

The result was usable.

Carrie's identity stayed consistent.

The main issue was motion quality. The walking movement felt slightly slow and floaty.

However, this was a motion problem, not a character consistency problem.

The character still looked like the same person.

Version B: Add a Side Profile Reference

The second version added:

@CHAR_SIDE

The full reference set became:

@CHAR_01_HERO
@CHAR_FULL
@CHAR_SIDE
@SCENE_01_SIDE

This version did not improve the workflow.

Instead, it introduced new problems.

During the clip:

• the background changed once
• the camera slowly moved away from the character

The shot no longer matched the original goal of a stable side-angle character shot.

This showed an important lesson:

More references do not automatically create more control.

An additional reference should solve a specific visual problem. If the current references already provide enough information, adding more inputs may introduce unnecessary changes.

Step 3: Continue With the Selected Reference Strategy

After comparing the two side-angle versions, I continued with the simpler reference setup.

For the final close shot, I used:

@CHAR_01_HERO
@CHAR_FULL
@SCENE_01_CLOSE

The goal was to see whether the same identity anchors could continue into a different framing.

The result worked well.

Carrie remained recognizable while the camera moved closer.

The face, hairstyle, outfit, and overall appearance stayed consistent with the earlier shots.

This created a simple three-shot sequence:

1. Medium shot
2. Side-angle shot
3. Close shot

The character changed position and framing, but the identity remained stable.

How to Evaluate Character Consistency Across Shots

A consistent character does not mean every frame looks identical.

The purpose is to keep the important identity signals while allowing the shot to change.

I use four checks when reviewing different shots.

Face Identity

Check:

• face shape
• facial proportions
• key facial features

The question is not:

Is this the exact same image?

The question is:

Would a viewer recognize this as the same character?

Hair and Appearance

Check:

• hairstyle
• hair color
• overall silhouette

Small changes in hair can make a character feel different, especially in close shots.

Outfit Consistency

Check:

• clothing type
• major colors
• important design details

Minor fabric changes are usually less important than the overall outfit identity.

Body Proportions

Check:

• height impression
• body shape
• overall silhouette

Full-body references are especially useful when the shot includes movement.

Common Mistakes When Using Ref2Video for Character Consistency

Adding More References Without a Clear Reason

More inputs do not always mean better results.

A reference should answer a specific question.

For example:

• Need facial identity? Use a clear face reference.
• Need body proportions? Use a full-body reference.
• Need a difficult angle? Consider an angle-specific reference.

Do not add references simply because they exist.

Changing Too Many Variables at Once

Character consistency becomes harder to evaluate when everything changes together.

Avoid changing:

• character
• environment
• camera
• lighting

in the same step.

A better approach is:

Keep the identity anchor stable.

Change one production variable at a time.

Confusing Motion Problems With Identity Problems

A video can have:

• good character consistency
• weak walking motion

or:

• good movement
• unstable identity

These are different problems.

In my side-angle test, the first version had slower movement, but the character identity remained stable.

The issue was motion quality, not reference quality.

Judging Only the First Frame

A character may look correct at the beginning but change during the clip.

Always review:

• the opening frame
• movement section
• final frame

Consistency is a video-level check, not only an image check.

FAQ

Should I use more reference images to keep AI characters consistent?

Not necessarily.

Start with the smallest reference set that provides the necessary identity information. Add more references only when a specific shot requires additional details.

What reference images are most useful for character consistency?

Different references solve different problems.

A face-focused image helps with identity.

A full-body image helps with proportions and outfit.

An angle-specific image can help when the shot reveals a view that previous references do not show clearly.

Can I create multiple AI video shots with the same character?

Yes.

The most reliable approach is to create a stable character reference system first, then reuse those references across different shots.

The goal is not to make every shot identical.

The goal is to keep the character recognizable while changing the scene, camera, and movement.

Does adding a side-profile reference always improve side shots?

No.

In this workflow, the side-profile reference did not improve the final result. It introduced unexpected background and camera changes.

The better choice was keeping the original identity references and using them consistently.

Final Takeaways

Building consistent AI characters is less about collecting more references and more about designing a controlled workflow.

Create a stable identity anchor.

Use it across shots.

Add new references only when they solve a real production problem.