A motion reference matters more than a long prompt
A text-only description can produce an ordinary dance in a room. The Window Dance joke depends on the camera position, timing and reaction. Our workflow combines a portrait, a scene preview and a motion reference.
Describe what changes and what stays
Example: Replace only the main dancer with the adult in my portrait. Keep the camera outside the window, the dancer inside the room, and the original sequence and reaction. Preserve recognizable facial features and natural anatomy. Keep the window frame, lighting and occlusion consistent. Do not add another person or text.
Keep the face readable
Use a clean portrait with enough facial detail. Tiny faces, sunglasses, profile-only views and strong beauty filters make identity harder to preserve. If the preview looks wrong, stop before approving the video.
Inspect the window and the ending
Check that the body remains behind the window, hands do not pass through solid frames, and the end still reads as noticing the camera. These are quality checks, not guarantees that a prompt can force perfect physics.
You do not need to write a prompt here
Window Dance AI includes the reference and instructions. Upload your portrait, check the preview and confirm the video step. For a different room, pet or custom choreography, this fixed workflow may not fit your goal.