NadouPro Motion Control Tutorial: Teach Static AI Characters Reference Actions

It is not hard to make a person in an image move. The hard part is making them move the way you want them to. Write a prompt describing a martial arts sequence, and the model gives you the average of what it imagines; ask for a street dance, and the result often looks like a generic exercise routine. In NadouPro's Video node, there is a mode designed specifically to solve this problem: Motion Control. The idea is straightforward—instead of describing the movement with words, show it directly.

How Motion Control Works: One Image Plus One Video

Motion Control takes two inputs: a character image and a reference video that contains the target action. The model's task is transfer—mapping the body movement, rhythm, and even expressions from the reference video onto your static character frame by frame, while preserving the character's appearance, clothing, and identity.

This is fundamentally different from ordinary Image to Video. Image to Video lets the model improvise freely; Motion Control follows the reference. The former depends on luck, while the latter depends on the reference you choose. Take a ten-second dance clip, run it three times with three different characters, and the motion fidelity remains stable while each character stays in place.

Kling 3.0 integrated into NadouPro is the main model for this type of task, with strong multimodal input capabilities and character motion physics and facial details that stand up to close viewing. For more complex scheduling, you can also use 3D Director for camera planning.

Complete Workflow

Using “make a traditional-style character dance modern dance” as an example, here is the workflow:

1. Prepare the character image: front view or three-quarter side view, full body in frame, clear lighting, and complete, readable facial and body structure. A character three-view sheet is the most reliable input

2. Prepare the motion reference video: one person on screen, complete unobstructed movement, clean background, and sufficient lighting. Dance and gesture materials are the most stable

3. Enter the Video node and select Motion Control mode: upload the character image and reference video

4. Write a short context description: explain the background, lighting, or atmosphere requirements. The movement itself is determined by the reference video; the text is only responsible for the environment

5. Generate and check: focus on whether the movement rhythm keeps up, whether the character's face remains stable, and whether the clothing drifts

There are two options for the background source: follow the environment in the reference video, or keep the scene from the character image. For comparison tests, it is very efficient to run the same dance with different scene backgrounds.

The Quality of the Reference Video Determines the Ceiling

Motion Control is sensitive to reference material quality. Here are three rules of thumb:

• A single subject is the most stable: footage with multiple people in the same frame can easily mix up movements, so crop it before uploading

• The start and end of the action should be complete: after transfer, truncated movement can feel awkward, as if it has no beginning or ending

• The body type of the reference should not differ too much from the final character: transferring a gymnast's flip onto a slightly heavier character can make the physics feel distorted

Who Is Using This Capability

Character animation creators: instead of animating keyframes frame by frame, they can feed in a live-action performance and make a concept character immediately able to act. Independent short film teams use it to turn design drafts into dynamic samples.

Brand mascot operators: mascots are usually static assets. Use Motion Control to make one dance, wave, or spin, and the material formats for social media content directly multiply.

Virtual human and AI host teams: one character asset paired with a motion library keeps identity from drifting during mass content production. The motion library can also be reused across projects—the same dance steps can be used for a realistic-style character today and an anime-style character tomorrow, with the style switching according to the character image.

Short drama and comic drama teams: for action-heavy scenes such as fight scenes and chases, using motion references instead of long prompt descriptions makes the director's intent much more accurately conveyed.

Motion Reuse: An Underestimated Lever for Scaling

Applying the same motion reference to different characters, different styles, and different backgrounds is the most economical use of Motion Control. One ten-second clip can generate dozens of outputs: realistic, illustrated, cyberpunk. For teams making serialized content, this spreads the one-time cost of choreography across the entire content matrix.

Take a typical scenario: a traditional Chinese-style IP team records a signature hand-gesture dance for the main character. After that, each solar-term poster video is generated based on this reference. The character style or scene changes with the season, but the gesture always remains the one the audience recognizes—this is how the memory point of a series can be fixed at low cost.

FAQ

Q: Can Motion Control and Multi-reference mode be used together?

They can be combined. Use Motion Control with the character image to lock identity and movement, then use Multi-reference input to supplement scene or prop information. This is suitable for shots with a large amount of information.

Q: Can multi-person actions be controlled?

In typical scenarios, single-person references produce the most stable results. For multi-person interaction shots, it is recommended to split them into single-person references and generate them separately, or use 3D Director for multi-character scheduling.

Q: Will the clothing from the reference video be carried over?

No. What is transferred is the movement, not the appearance. The character's clothing follows the character image you upload.

Q: Will hand details become blurry?

Hands are a difficult part of motion transfer. Results are best when the gestures in the reference video are clear and complete; for close-up hand shots, it is recommended to generate them separately and then use Frame Interpolation / Ultra HD to enhance them.

Q: How long can a reference clip be at most?

Based on the current capabilities of mainstream models, reference clips from a few seconds to a little over ten seconds produce the most stable transfer quality. For longer actions, it is recommended to process them in segments and then stitch them together in Clip.

For more features, visit NadouPro: NadouPro

Contains AI-generated information that may be incorrect. Please verify independently.