Acting Reference Video

✅ DO ✅ ❌ DON’T ❌
The camera is stable, still, and at actor’s level. Don’t zoom, move the camera, or use tilted angles.
Record only one actor per video. Avoid videos where multiple humans are in the foreground.
Use fitted clothing so the model can clearly track the actor’s limbs. Don’t use loose clothing that hides or obscures the actor’s limbs.
Make sure the actor is standing on the ground at the first frame of the video. Don’t start the video with the actor jumping, mid-air, or in an unclear stance.

Follow the rules above for the best motion capture. Our model tracks full body motion (including head and fingers), but not facial expressions or objects.

Example of an ideal video:

ferocity.mp4

First frame / Image reference

  1. Your input image must feature a human-like character with clearly visible limbs (arms, legs, neck, etc.). Characters with unusual proportions (e.g., very long legs, very short arms) may be harder to animate accurately.
  2. Use a landscape format, ideally 16:9. Other formats are supported, but 16:9 images deliver the most reliable results. The final output will always be a 16:9 video, regardless of input aspect ratio.
  3. Ensure the ground in your input image is relatively flat.
  4. Align the character’s orientation and posture as closely as possible with the actor’s orientation and posture in your video.

Example of an ideal acting & image references pair:

regrets after murder - knife version.png

murder acting.gif

Our recommendation: To ensure the character is in the right position, you can use Nano Banana and use the first frame of the acting video as a reference.

Prompt

To get the best results, use your prompt to clarify anything the model can’t infer directly from the image or acting video.