How to Choose a Motion Transfer Reference Video: A 60-Second Check

Motion transfer carries over the movement, timing, subject positions, camera work, and edits from a reference video. If the reference is a poor fit, a longer prompt usually cannot repair the underlying performance.

Before generating, run each candidate clip through the five checks below. It takes about a minute and can prevent wasted test runs.

1. Does the movement already match your goal?

Ignore the people in the clip for a moment and watch only the performance. Are the sequence, speed, and intensity already close to the result you want?

Choose a different reference if you need different choreography, slower timing, or different camera work. Character images supply appearance; they do not redesign the full performance in the reference video.

For a practical check, choose a ready-made template or upload a candidate clip in Motion Transfer. Before generating, play it from beginning to end and judge only the movement and camera work. Then review the character images and prompt.

2. Can you clearly see the important movement?

Pay close attention to faces, hands, feet, and physical contact. Fast turns, hands covering faces, and people crossing in front of one another force the model to infer missing information.

Prefer clips where the subject stays visible, the limbs remain mostly inside the frame, and each movement is easy to follow. When a complex action is unavoidable, test the shortest complete segment first.

3. Do the number and positions of people remain clear?

Solo clips are usually the simplest to control. A two-person clip needs two clearly distinguishable performers. A crowd scene also needs a stable lead, headcount, formation, and depth order.

For a two-person or crowd clip, answer three questions:

• Who is each person at the beginning?
• Can you still identify them after an obstruction or crossing?
• Are the headcount and positions clear after every cut?

If these answers are unclear while watching the original clip, the generated version is more likely to swap or duplicate people.

4. Is the camera work worth preserving?

Pushes, pans, zooms, framing, and cuts from the reference usually carry into the result. Make sure those choices already support the video you want.

Rapid cuts, sudden zooms, and frequent changes in shot size add difficulty. For the first test, start with one shot and one complete action. Move to a longer sequence after the characters remain stable.

5. Can the hardest frame work?

Do not judge only the opening. Find the most difficult moment in the entire clip, such as:

• the first side or back view;
• a hand passing across a face;
• two people blocking each other;
• the lead entering a crowd;
• the first frame after a cut.

Your character references should provide the appearance information needed at those moments. A front-facing portrait may need a matching side or full-body image when the video contains many turns or full-body actions.

A simple clip scorecard

Check each candidate against this table:

Check	Passing condition
Movement	The performance already matches the goal before replacing anyone
Visibility	Faces, hands, feet, and important contact remain mostly clear
People	The headcount stays stable and each person is easy to distinguish
Camera	The framing, camera movement, and cuts are worth preserving
Length	You can begin with a short, complete action
Rights	You have permission to use and publish the footage

If a clip fails more than one of these checks, replacing it will often save more time than repeatedly rewriting the prompt.

Final check before generating

1. The reference video already contains the movement and camera work you want.
2. Each character image has a clear role.
3. The images cover the necessary front, side, or full-body views.
4. You have identified the hardest turn, obstruction, and contact point.
5. The first test uses the shortest complete segment.

Resolve movement, visibility, and subject relationships while choosing the clip. Use the prompt to describe clear assignments and constraints. Each test will then be easier to compare, and the next change will be easier to choose.