A reference action video is the motion instruction for AI motion transfer. The clearer the reference, the easier it is for the AI to recreate the movement on your portrait subject.
The best reference videos are not always the most cinematic. They are readable. They show one subject, clear body movement, stable framing, and enough visual information for the model to understand the action.
Use these seven tips before generating a video in Animaker Dev.
1. Keep One Main Subject in Frame
AI motion transfer works best when the reference video has one clear performer. If the video contains multiple people, overlapping limbs, or crowded background movement, the model may struggle to identify which motion matters.
Good reference:
- One performer.
- Body clearly visible.
- No other person blocking the action.
- Simple background.
Risky reference:
- Group dance.
- Crowd scene.
- Person walking behind the subject.
- Fast cuts between performers.
If you want to use a group dance, crop or choose a segment where one person is visually dominant.
2. Use Stable Framing
Stable framing makes the action easier to read. A handheld video with heavy shake can create confusing motion because the camera movement mixes with the body movement.
Stable does not mean boring. The performer can move. The camera just should not add too much chaos.
Best practice:
- Keep the subject centered.
- Avoid sudden zooms.
- Avoid fast pans.
- Keep the body inside the frame.
- Use a short clip with continuous motion.
3. Show the Body Parts That Matter
If the action uses hands, arms, legs, or torso movement, those parts should be visible in the reference video. Hidden hands and cropped feet are common reasons for weak AI motion transfer results.
Use this quick rule: if a human cannot clearly see the movement, the AI may not follow it well either.
| Action Type | Important Visibility |
|---|---|
| Dance | Full body, arms, legs, feet |
| Hand gesture | Hands, wrists, face |
| Walking | Legs, feet, torso direction |
| Presenter motion | Face, hands, upper body |
| Product demo | Hands and product interaction |
4. Avoid Props That Hide the Motion
Props can be useful in real video, but they often make AI motion transfer harder. A bag, microphone, phone, cup, or large sleeve can hide hands and confuse body structure.
If the prop is essential, use a simple one and keep it visible. If the goal is a clean first result, choose a reference without props.
Also check the portrait photo. If the person in the photo is holding an object but the reference subject has empty hands, the mismatch can create distortions.
5. Reduce Motion Blur
Fast movement can look exciting, but heavy blur removes detail. Arms, fingers, feet, and face direction become harder to track. That can lead to distorted frames or weaker motion.
To reduce blur when shooting your own reference:
- Use better lighting.
- Avoid very low-light rooms.
- Keep the camera steady.
- Choose a shorter, cleaner action.
- Avoid extremely fast spins for the first test.
You can still test energetic movement. Just start with a readable version first.
6. Match the Reference to the Portrait
The reference video and portrait photo do not need to be identical, but they should not conflict too much. If the reference action is full-body dancing, a full-body or head-to-waist portrait usually gives better context than a face-only selfie.
Check these pairings:
| Portrait Photo | Reference Video | Expected Difficulty |
|---|---|---|
| Full-body standing photo | Standing dance clip | Easier |
| Head-to-waist portrait | Upper-body gesture | Easier |
| Tight face-only selfie | Full-body dance | Harder |
| Seated portrait | Running or jumping clip | Harder |
| Hidden hands | Hand-heavy gesture | Harder |
Better matching does not guarantee perfection, but it reduces unnecessary conflict.
7. Start with a Short, Simple Clip
For your first generation, choose a short action clip with one clear movement idea. You can always test longer or more complex references later.
A good first reference clip might be:
- A short dance move.
- A simple wave and pose.
- A walk toward camera.
- A product presenter gesture.
- A clean fashion turn.
This gives you a baseline. Once you see how your portrait responds to the motion, you can experiment with faster choreography or more expressive clips.
Common Reference Video Mistakes
| Mistake | Why It Hurts | Better Choice |
|---|---|---|
| Multiple dancers | The AI may mix body motion | Use one clear performer. |
| Fast camera shake | Camera motion hides body motion | Use stable framing. |
| Cropped hands or feet | Key movement disappears | Keep the body in frame. |
| Dark video | Details are lost | Use brighter lighting. |
| Props in hands | Body structure is hidden | Use empty hands for first tests. |
| Long chaotic clip | Too much action to interpret | Use a short readable segment. |
How to Test Improvements
When a result looks weak, change only one variable. If you replace both the portrait and the reference video at the same time, you will not know which change helped.
Try this order:
- Keep the photo and simplify the reference video.
- Keep the reference video and use a clearer portrait.
- Shorten the reference clip.
- Remove props or crowded movement.
- Try a pose that better matches the photo.
Generate with Animaker Dev
When your reference clip is ready, upload it with a clear portrait photo in Animaker Dev. The task result is saved in your dashboard, and failed tasks refund the consumed credit automatically.
