How to Do the AI Dance Trend With Your Own Face (Step by Step)
A working tutorial for copying a dance from a Reel onto yourself: which clip to pick, the one parameter that decides whether identity survives, the reference photos to supply, and what a finished clip costs.
This is the tutorial, not the model round-up. One model, one configuration, in the order you actually do it.
Step 1 — pick the right source clip
Not every dance survives motion transfer. What works:
- 5–8 seconds. You can feed a reference clip of 3–30 seconds, but identity drift compounds with length. Pick the recognisable section, not the whole song.
- One person in frame. Two dancers means the model has to decide which body is yours.
- Full body visible, stable camera. Rapid cuts and whip pans give the model contradictory motion to follow.
- Feet in frame if the dance uses them. Cropped legs get invented.
Step 2 — take the reference photos properly
This is where most attempts are lost before any money is spent. A dance rotates your head through angles a selfie never covers.
Supply a frontal shot, both three-quarter angles and a profile, all in similar lighting, whole head in frame, no sunglasses, no heavy shadow across one side. If the outfit matters, shoot the photos in it — the model reads wardrobe from your references, not from the source dancer.
Step 3 — the setting that decides everything
Use fal-ai/kling-video/v3/pro/motion-control at $0.168 per output second (read 2 August 2026).
Set character_orientation to video.
This is not a stylistic choice. On video, the reference clip drives the framing and the model honours the elements entry that binds your face to the character. On image, the framing follows your still and that face binding does nothing — the request succeeds, you are billed, and the identity drifts. If you have tried this before and concluded it does not work, this parameter is the first thing to check. The 30-second ceiling also only applies in the video orientation.
Use Pro rather than Standard whenever the face is on screen. Standard is $0.126/s, Pro is $0.168/s, and in our side-by-side runs Pro held identity noticeably better. On an 8-second clip that is $1.34 against $1.01 — a third of a dollar to avoid a re-run.
Step 4 — check it before you post
Watch for the three tells: the face softening in the second half, wardrobe changing mid-clip, and hands. If any of those show up, shorten the clip rather than re-rolling the same length.
fal auto-refunds failed motion-control jobs, so a rejected generation is not billed — a retry after a policy rejection costs nothing but time.
What it costs
An 8-second clip on Pro: $1.34. Reference photos: free. Most of your spend on a first attempt goes into learning that the source clip was a bad choice, which is why step 1 is step 1.
What will not work
Feeding a real face to Seedance or Veo — both refuse real human likenesses, and the filter sits upstream of every reseller. Face-swap apps in the $10–15/month tier do face-only compositing, typically 720p and capped near 15 seconds; on a dance, where the body moves and the head turns, the seam shows exactly when the motion gets good.