The 24K Magic AI dance is a 10-second TikTok edit that puts you, from one photo, on a sunny cruise-ship deck dancing to Bruno Mars's "24K Magic". It recreates the Hotel Transylvania 3 deck scene, and on videosi you make it without writing a prompt.
This guide goes deeper than the FAQ on the trend page. It covers where the dance came from, what the CapCut template is, why faces drift, how labels and rights work, and whether a general video model can do the same job.
Key takeaways
- The dance recreates the 2018 Hotel Transylvania 3 cruise deck scene, set to Bruno Mars's 2016 song "24K Magic".
- The AI version started spreading on TikTok around 31 October 2026; who posted first is unconfirmed.
- The clip is 10 seconds with five beats, from a grin close-up to a low-angle feet shot and wide deck shots.
- videosi makes it from one photo with no prompt; a standard 720p video is 9 credits, with packs from $6.99 for 10 credits.
- Likeness drifts most in wide shots, so use a clear front-facing photo.
In this guide
- Where does the 24K Magic dance come from?
- What is the Hotel Transylvania 3 scene people are copying?
- What happens second by second in the 10-second clip?
- What is the CapCut template everyone mentions?
- Why doesn't the dancer look like me?
- Do I have to label the video as AI on TikTok?
- Can I put a celebrity or a cartoon character on the deck?
- Is the song copyrighted, and does the clip include it?
- How do I make it on a phone?
- Can I do this myself with Sora, Veo or another video model?
- What does it cost?
Where does the 24K Magic dance come from?
The moves are not new and not AI. The dance recreates the deck scene from Hotel Transylvania 3: Summer Vacation, the 2018 Sony Pictures Animation film, set to "24K Magic", the 2016 single by Bruno Mars. Anyone can open the film's Wikipedia page and the song's Wikipedia page to confirm the basics.
What is new is the AI layer. In late October 2026 TikTok filled with near-identical posts that swap a new person onto the deck. The earliest posts we found with the shared caption are dated 31 October 2026. We could not confirm who posted first, and a template site that names an account and a date of 12 October 2026 is unverified. Treat any claim about the creator as unconfirmed, and any view count as unconfirmed too.
What is the Hotel Transylvania 3 scene people are copying?
On the trend page the scene is described as the "Looking Good Dracula" deck dance on the cruise ship in the film. In the film the ship is where the monster family spends its vacation, which is why the setting is a bright deck with white railings, a staircase and a blue sky. The AI clip borrows that look: a dim wood-panelled cabin close-up, then the sunny deck.
The camera work is part of the appeal. The clip opens on a grin with both hands tucked under the chin, cuts to a wide shot in the middle of the deck, drops to a low-angle shot with the feet right at the lens, then returns to wide shots beside the stairs. The 24K Magic AI dance page keeps all four setups, because they are what make a viewer recognise the scene in the first second.
Older, non-AI recreations exist too. In 2022 a TikTok user posted the film scene beside his own recreation and reached 1.1 million views, so the dance was a known meme years before the AI version.
What happens second by second in the 10-second clip?
The clip is built from five beats. From 0 to 2 seconds you see the grin close-up in the cabin. From 2 to 4 seconds the shot cuts to the deck, with you in the middle starting to shuffle. From 4 to 5.5 seconds the low-angle feet shot shows your shoes huge in the foreground, sliding and stepping.
From 5.5 to 8 seconds the wide shot beside the staircase shows bent-knee squats, a kick and loose arm swings. The last two seconds repeat that wide shot as the dance gets faster and ends mid-move on the beat. Because the timing is fixed by a motion reference, the cuts land where the music expects them.
What is the CapCut template everyone mentions?
Most of the early posts share one caption, along the lines of "How I made the viral 24K Magic AI dance trend", and tell viewers to open a CapCut template and drop in a photo. That is a template, not a TikTok filter, and templates change, move or disappear without notice. We do not know who maintains the one in those posts.
The practical difference is where the work happens. A template runs inside an app on your phone and depends on a link surviving. The videosi version is a web page: you upload one photo, choose a format and download a 10-second MP4. videosi is independent and not affiliated with TikTok, CapCut or Bruno Mars.
Why doesn't the dancer look like me?
Likeness is the most common complaint with any one-photo dance, and it is rarely random. The usual causes are sunglasses, a side angle, a group photo or a dark, grainy picture. The model needs a clear, front-facing view of your face to hold it across four different camera setups.
There is also a built-in difficulty in this routine. In the wide deck shots your face takes up only a few pixels, so it drifts most there. The close-up at the start is easier to hold, and the prompt includes a face-shape rule so the opening shot does not slim or reshape you toward the original dancer. Use a front-facing, waist-up or full-body photo in soft light, and expect some drift in the wide shots. Likeness is never guaranteed.
Do I have to label the video as AI on TikTok?
Check the current rules in the app, because platforms update them. TikTok has offered an AI-generated content label for realistic synthetic video, and many creators in this trend simply say so in the caption. A video of a real person dancing where they never danced is exactly the kind of clip a label is meant for, so adding one is the safe habit.
It also helps with trust. Viewers who see a familiar scene and a famous song are usually in on the joke, and a plain "AI dance, made from one photo" caption avoids arguments in the comments. We cannot speak for what the platform enforces, so treat this as practical advice and not as a policy statement.
Can I put a celebrity or a cartoon character on the deck?
On TikTok the trend includes footballers, tennis and basketball players, actors and cartoon characters. videosi does not accept photos of celebrities, public figures or minors, so those versions are not possible here. Only an adult who has agreed to it should be in the photo, and that includes you.
The page also never uses the singer's or a film character's face. The dancer is whoever is in your photo, in their own clothes. A photo of your own pet can sometimes work, but the dance was made for a person and pet likeness is not guaranteed.
Is the song copyrighted, and does the clip include it?
"24K Magic" is a commercial recording, so the rights belong to its owners and not to us or to you. The videosi clip comes with the trend sound already on it, taken from the same reference as the motion. We have not matched that audio against the official track, and TikTok lists most clips as an original sound cut from the song.
For private sharing and normal posting on the platform where the sound is offered, that is how the trend works. If you plan to use the video in an advertisement or on a channel you monetise, check the licence yourself first. This is not legal advice.
How do I make it on a phone?
There is no app to install. Open the 24K Magic AI dance page in your mobile browser, upload one photo from your camera roll, pick 9:16 for TikTok, Reels or Shorts, and press Generate. Rendering takes several minutes, so you can leave the tab and come back.
When it finishes, download the MP4 and post it from your gallery. The format options are 9:16, 1:1 and 16:9. If the result is off, the fix is almost always a better photo, not a longer wait.
Can I do this myself with Sora, Veo or another video model?
You can try, and the honest answer is that it is harder than it looks. The page lists the pitfalls: typing the singer's name into a prompt is blocked by most tools, and when it works you get a celebrity look-alike you should not post. Without a motion reference, the close-up, the feet at the lens and the stair-side shots come out as a generic dance on the spot.
General models also tend to keep the original dancer's face or clothes, or slim your face in the close-up, unless the replace instruction and the identity lock come first. Each attempt costs model credits and several minutes, and getting four camera setups right in 10 seconds usually takes several tries. videosi uses ByteDance's Seedance with a built-in prompt and a 10-second motion reference, so you skip that tuning. The full prompt stays unpublished.
What does it cost?
There is no free generation and no free trial. A standard 10-second video at 720p is 9 credits, and better models or higher resolutions cost more. Credit packs start at $6.99 for 10 credits and never expire, plans start at $29.90 a month, and a failed render returns its credits.
Since the trend is young and its stage is still emerging, the best time to post is now. Make yours from the 24K Magic AI dance page while the sound is still circulating.
