The Smooth Criminal AI dance is a 15-second AI video where one person from a single photo tips a white fedora and leads a club of suited dancers into the famous forward lean. This guide answers the deeper questions behind it: how the lean works, what the AI version changes, which photos work and what to expect when you post it.
Key takeaways
- The lean is a 45 degree forward tilt with a straight back; the original video used cables and a harness, and a 1993 patent covered stage shoes for live shows.
- The Smooth Criminal AI dance is a 15-second club scene with a hat tip, an overhead lean and a groove, made from one photo.
- It adds only a white fedora; your face, hair and clothes come from the photo, and the other dancers are AI extras.
- Your face is weakest in the overhead lean, so judge the clip by the opening and the final groove.
- Platforms may mute the song, so keep the AI label on and swap in the app's own track if needed.
In this guide
- How does the Smooth Criminal lean work in real life?
- Why is the AI version a club full of dancers instead of one person?
- Why does the video add a white fedora?
- How is this different from the Michael Jackson AI dance page?
- What kind of photo works best?
- Why does my face look weaker in the overhead shot?
- What happens second by second in the 15 seconds?
- Where does the Smooth Criminal lean come from?
- Can I post it on TikTok, and what about the music?
- Do I need to label it as AI?
- What does it cost and what happens if a render fails?
How does the Smooth Criminal lean work in real life?
The lean is a forward tilt of about 45 degrees with a straight back and flat feet, held for a moment and then released. On Wikipedia's Smooth Criminal entry, the effect in the original video is credited to cables and a harness, because the body's center of mass moves further forward than any unaided dancer can support.
For live shows, Michael Jackson's team took a different route. US patent 5,255,452, titled Method and means for creating anti-gravity illusion, names Michael J. Jackson, Michael L. Bush and Dennis Tompkins as inventors. It was filed on 29 June 1992 and granted on 26 October 1993. It describes footwear with a heel recess and a V-shaped slot that hooks onto a post rising through the stage floor, so the performer can lean past their balance point and then slide free to move normally.
That history matters for the AI version. A video model has no harness and no pegs. It has to imitate the silhouette of the lean from a motion reference, which is why the shape of the bodies in the overhead shot is the part people judge first.
Why is the AI version a club full of dancers instead of one person?
The trend is built around the forward lean, and the lean reads best as a group move. In the Smooth Criminal AI dance, a high overhead shot shows the whole floor tipping forward together between about 4 and 6 seconds, with straight bodies, planted heels and long shadows under the spotlight. One person leaning alone looks like a stumble; a line of suited dancers leaning together looks like choreography.
The structure is also what makes it easy to share. The first two seconds give a face and a hat, the overhead shot gives the spectacle, and the last seven seconds give a groove you can loop. Each beat has one job, so the clip works in a feed even with the sound off.
Why does the video add a white fedora?
The fedora is the one prop that stays the same in every clip. In the trend it opens the video: the lead lowers a white fedora with a black band and looks into the lens, and then the dancers behind touch their own hat brims. It is the cue that tells a viewer which song and which scene this is before any music registers.
On videosi the hat is added to your photo's subject and nothing else is. Your face, hair and clothes come from the picture. There is no white suit, no glove and no celebrity face, so the result reads as you wearing the hat rather than as an impression of someone else. If your photo already shows a hat, the model has two hats to reconcile, which is one of the common reasons a render drifts. Pick a photo with a bare head.
How is this different from the Michael Jackson AI dance page?
They are two separate videos with different scenes. The Michael Jackson AI dance is a solo routine set to Dangerous, in your own clothes, with no hat and no crowd. The Smooth Criminal AI dance is a club scene: close-up hat tip, overhead lean, wide groove, with a floor of AI dancers around you.
If you want to be the only person on screen, take the solo page. If you want the lean and the group shot, take this one. Both are 15 seconds and both come with their own sound already on the finished MP4, so you do not need to add a track for either.
What kind of photo works best?
Use one clear, front-facing photo of one adult who agreed to be in it. Good light on the face matters more than resolution. A covered or blurry face, sunglasses, a hat, or a second person in the frame are the usual causes of a result that does not look like you.
Clothes carry over too. In the clips we watched, people appear in their own shirts, suits and jewellery, so a photo from the waist up in something you like wearing gives the model more to keep. Avoid heavy filters, because the model will treat the filtered skin as the real face. The video is built for one person, so pets and group photos are not a fit.
Why does my face look weaker in the overhead shot?
Because you are small in that frame. The overhead lean is a wide, high-angle view of a whole dance floor, and the lead is one figure among many. A video model spends its detail where the subject is large, so the close-up at the start and the wide groove at the end carry more of your likeness than the lean does.
This is a limit of the format, not a fault in your photo. The practical advice is to judge the clip by the opening hat tip and the final groove, where your face fills more pixels. The other dancers in the club are AI extras, not you or your friends, and their faces change from take to take. Likeness is not guaranteed, and a failed, timed-out or blocked render returns its credits automatically.
What happens second by second in the 15 seconds?
There are five beats. From 0 to 2 seconds you get the hat tip in warm club light. From 2 to 4 seconds the camera cuts to a high overhead shot of the wooden floor while the dancers touch their hat brims. From 4 to 6 seconds the whole floor leans far forward. From 6 to 8 seconds the dancers rise and the lead squares up to the camera. From 8 to 15 seconds a wide shot from floor level shows the lead grooving with kicks and side steps at the front of the line.
Knowing the timing helps you plan a post. The lean lands about a third of the way in, so a caption or a text hook should appear in the first two seconds, before the cut to the overhead view.
Where does the Smooth Criminal lean come from?
It comes from the Smooth Criminal sequence in the 1988 anthology film Moonwalker, which Wikipedia describes as a musical film that strings together short films, several of them long-form videos for songs from the 1987 album Bad. Smooth Criminal was the seventh single from that album and reached number seven on the Billboard Hot 100, according to the song's entry.
Our own trend notes say the earliest AI clip in this format we found on TikTok is dated 4 September 2026, and that the most-watched one we found has about 158,000 views. Nobody has confirmed who made the first one or which tool it came from, and we do not name a creator. Our trend radar flagged the phrase as budding on 1 November 2026, so the trend is early rather than established.
Can I post it on TikTok, and what about the music?
Yes, you can upload the MP4 to TikTok, Reels or Shorts. The song belongs to its rights holders, so a platform may mute or limit the audio on your clip. If that happens, remove the sound and add the same track from the platform's own music library, which is licensed for use inside that app.
Our clip carries the same 15-second instrumental part of the recording that the AI posts we checked used, without vocals. That is useful for timing, because the lean lines up with the instrumental build. videosi is independent and not affiliated with Michael Jackson, his estate or the song's rights holders.
Do I need to label it as AI?
Yes, turn on the AI-generated label when you post. It costs you nothing, and it protects the clip from being reported as a deceptive edit. The video shows a realistic person in a realistic place doing something they never did, which is exactly what the labels are for.
It is also the honest framing. The dancers around you are generated, the club is generated, and the lean is a model's imitation of a motion reference. Saying so tends to help rather than hurt, because the trend itself is the AI part.
What does it cost and what happens if a render fails?
There is no free generation and no free trial. A standard 15-second video on the Standard model at 720p costs 13 credits. The Single pack is $6.99 for 10 credits, so plan on more than one pack if you want to try a couple of photos. Better models and higher resolutions cost more.
A failed render returns its credits automatically, and there are no refunds beyond that. Because the page writes the prompt and holds the motion reference for you, the only choices you make are the photo and the format: 9:16, 1:1 or 16:9. The page also shows the structure of the written prompt, for example that it opens with the instruction to “Replace the lead dancer in Video 1, the man with glasses in the white fedora with a black band, pink gingham shirt and grey trousers”, but the full prompt is not published.
Ready to try it? Upload one photo on the Smooth Criminal AI dance page and press Generate.
