Choose a clearer motion reference
Written by Kadenly · Updated
Prepare a character image and motion reference with clear framing and visibility. Learn input checks and the limits of our current motion transfer preview.
Purpose and limitations
Reference-video Motion Transfer is not currently available. These checks help prepare future inputs. The available Image to Video workflow can suggest a small action from a still image, but cannot reproduce an uploaded video’s exact motion.
Start with a single performer
Pick a short shot with one visible performer and a stable camera. Frequent cuts and people crossing the frame make the intended movement ambiguous.
Match the character framing
Pair a full-body performance with a character image that shows the body clearly. Keep the body orientation reasonably similar when possible.
Keep the first direction simple
Prioritize the reference movement before requesting large scene changes. Review the framing and body motion before adjusting the background.
English prompt example
This English example describes a small action for Image to Video. It is not a motion-transfer request and does not prove reference tracking or exact movement reproduction.
The character makes a small, slow arm gesture while standing in place. Keep the full body visible, the camera locked, and the background steady. One continuous shot.Input and result checklist
- Choose one performer in a single uninterrupted shot.
- Keep the full body visible if the movement involves arms or legs.
- Avoid cuts, heavy occlusion, rapid camera movement, and limbs leaving the frame.
- Match the character reference framing and orientation to the performer.
- Confirm a compatible video-input model before treating the reference as usable; current generation does not accept this workflow.
Common issues and adjustments
The performer becomes obscured
Choose a reference with a clean foreground and no crossing people. Hidden limbs make the intended motion ambiguous.
Character and reference framing do not match
Use a full-body character image for a full-body performance. A face close-up does not describe body structure or clothing.
The output does not follow the reference
The current still-image workflow does not track a reference video. A prompt may describe an action but is not a replacement for a compatible motion-transfer model.
Sources and related policies
These are original preparation tips and prompt examples from our team. Model capabilities change; check current settings before generating. For input requirements and usage rules, see:
This uses a single reference image, not two-character swapping or reference-video motion transfer.
Try a small motion in Image to Video