Quick answer
Starrd's Smooth Criminal Remix template turns two photos into Cadel and Mia's viral Smooth Criminal remix dance. Photo 1 dances on the left and photo 2 on the right, side by side, full figure, on a locked-off phone camera, doing the whole routine on the beat: a deadpan start with a head snap, both hiding their faces, a hair whip, punches, an open-mouthed laugh, a cheeky face and arms up at the end. Instead of the original rooftop, the dance happens in the place from your photos: the template picks whichever photo shows more of a real place (a bedroom, a living room, a street) and opens it up into a dance floor. Everyone keeps their own face, hair and outfit, and pets dance upright with real paws. The original sound is muxed on top. 15 seconds, vertical 9:16, Seedance 2.5, 480p, 150 credits (about $7.50).
The same dance, any subject — tap one to make it with your photo.
Two photos in, and out comes you and a friend (or your dog) doing the viral Smooth Criminal remix dance: every move on the beat, every face on cue, in the place from your own photo. Here it is:
Smooth Criminal Remix
Two photos. You and a friend (or your pet) do the viral Smooth Criminal remix dance in the place from your photo. 15 seconds.
What You Get
- The whole routine. All 15 seconds of Cadel and Mia's dance, side by side, full figure, one continuous take.
- On the beat. The remix runs at about 100 BPM, and every duck, punch, point and arm swing lands where the original's does.
- The reactions. Deadpan, hiding, the hair whip, the laugh, the cheeky face and the arms-up finish, on their moments.
- Your place. The dance happens in the room or street from your photo, not on someone else's rooftop.
- Your own outfits. Faces, hair and the clothes from your photos, down to the shoes.
- People or pets. A dog dances upright with real paws.
- 15 seconds, vertical 9:16, with the original sound, ready for TikTok, Reels and Shorts.
What the Video Actually Is
Cadel and Mia posted the clip from @cadelandmia on August 4, 2026, captioned "SMOOTH CRIMINAL REMIX". It passed 8.2 million plays, 680,000 likes and 76,000 saves. The two stand side by side on a rooftop at sunset in front of a phone on a tripod and do a short, punchy routine to a remix of "Smooth Criminal" ("you've been hit by, you've been struck by…"), with as much face as footwork.
Why it suits AI: it's a two-person dance that needs two people in the same place at the same time, plus the confidence to do it. Two photos solve both, and putting it in your own living room instead of their rooftop makes it look like you actually filmed it.
The 15 Seconds, Beat by Beat
| Time | The dance | The faces |
|---|---|---|
| 0:00 | Side by side, arms down | Deadpan; the right dancer's head snaps left and right |
| 2.3s | Both duck with bent knees | Faces hidden under both hands |
| 3.5s | Left dancer whips the hair round; right dancer puts both arms behind the head | Looking off to the side |
| 4.6s | Both snap back to the lens | The right dancer mouths "struck by" |
| 5.3s | Fists up, punching on the beat | Grins; the left dancer throws the head back laughing at 6.8s |
| 8.6s | Left dancer points both fingers up; right dancer puts a hand on the head | A cheeky face |
| 10.0s | Fists up, shimmying | Grinning at the lens |
| 12.5s | Grooving side to side | Laughing |
| 14.4s | Arms up high | Laughing |
Your Look, Your Place
- Your own faces and outfits. Nobody gets the original's crop top, jersey or baggy jeans.
- Your own place. Upload a photo with a real background, like your living room, your bedroom or your street, and that's where the dance happens. If you both send plain-wall selfies, you dance in a simple room in that wall's colour.
- Your own sides. Photo 1 always dances on the left, photo 2 on the right.
Put the photo with the best background in either slot: the template checks both and uses the one that shows more of a place. Full-length photos give the best outfits, because the dance shows you head to toe the whole time.
The Fastest Way
Smooth Criminal Remix
Two photos, the whole routine on the beat with the original sound, in the place from your photo.
Two photos, 150 credits (about $7.50), no prompt. It renders on Seedance 2.5 at 480p, vertical 9:16, 15 seconds.
Build It Yourself
1. Turn the dance into a depth map, then cut the place out of it. Run the clip through a depth-estimation model (Depth Anything V2 works). Then use a segmentation model (SegFormer trained on ADE20K works) to keep only the two dancers. Replace everything else with a plain backdrop and one clean floor gradient: the camera never moves, so a single floor works for every frame. Now the depth map carries the dance and nothing about the rooftop, and the place can come from your photo.
2. Pick the place. Look at both photos and choose the one with more of a real place behind the person. Describe that place in one sentence (walls, furniture, colours, light) and open it up so there's floor in front of it to dance on.
3. Make one storyboard sheet of six moments. The deadpan start, the duck-and-cover, the hair whip, the laugh, the cheeky face and the arms-up finish, generated from both photos in that place, with a depth-map sheet of the original poses as a layout guide.
A photorealistic still from a dance video filmed on a phone on a tripod, both dancers full figure side by side in a cosy living room with a stone fireplace and wood floors: the person from reference image 1 on the left and the person from reference image 2 on the right both duck with bent knees and hide their faces under both hands, each wearing exactly the clothes from their own photo. Natural indoor light, a touch of grain. No text.
4. Drive it with the depth map and the song. Feed the sheet, both photos and the depth video (with the song inside it) to a model that copies motion from a reference video (we recommend Seedance 2.5). Then write what it can't read from a depth map: who dances on which side, that the camera never moves, and a timed list of the facial reactions. Render it silent and lay the original audio back on at the end.
Common Mistakes
Leaving the background in the depth map. The rooftop wall and the horizon come through as shapes, and your living room grows a parapet. Keep only the dancers and a plain floor.
Forgetting the faces. The dancers' faces are tiny in a full-figure shot, so the depth map carries none of the expressions. If you don't time the reactions in the prompt, you get two blank faces doing a comedy dance.
Letting the model move the camera. The original is locked off on a tripod. Say so, or the model adds a slow push-in that breaks the phone-video look.
Swapping sides. Without "photo 1 on the left, always", the model sometimes puts the taller person on the left.
FAQ
How many photos? Exactly two: photo 1 on the left, photo 2 on the right.
How long is it? 15 seconds, vertical 9:16.
What does it cost? 150 credits, about $7.50.
Where does it happen? In the place from whichever of your photos shows more of one.
Related Reading
- How to Make the 'APT.' AI Video: you and a friend in the ROSÉ and Bruno Mars video
- How to Make the 'ATEEZ BAD' AI Dance Video: three photos lead the BAD dance on stage
- How to Make the "What's Up In A Minute" AI Video: the swiss side quest trio dance
- How to Make Your Pet Dance (AI Video): every way to get your dog or cat dancing
- Viral AI Video Trends 2026: every trend we've turned into a template this year
