The that boy that boy dance AI trend is a nine second car-seat routine in which a person points away twice, jabs a fist at the camera on "sus", winks, cups an ear and waves it all off. On videosi you make it from one photo of yourself, press Generate and download an MP4 that comes with the trend sound already on it. It is a young trend, so most of what follows is about how to make a good version before everyone else does.
Key takeaways
- "Sus" is slang for suspicious, so the whole routine is a call-out joke: you spot the shady one and point at the lens.
- The earliest car-seat version we found, on 15 October 2026, passed 20 million views, but the song behind it is not confirmed.
- The finished video has the trend sound on it, yet the mouth moves loosely and does not lip-sync the words exactly.
- A standard nine second video (Standard model, 720p) costs 8 credits, with no free generation, and a failed render returns its credits.
- Only photos of adults who agreed are accepted, never celebrities or anyone under 18.
In this guide
- Why does the joke only work with the word "sus"?
- Is this the same trend as the 2020 standing dance?
- Which song is it, and why do we not name one?
- What does each gesture in the routine actually say?
- Why does the mouth not match the words exactly?
- Should I pick "Just you" or the caricature look?
- What makes a photo work for a hand-heavy clip?
- Why do hands in front of the face cause problems?
- Is it fair to make this about a friend?
- What does one try cost, and what if it fails?
- Is it worth posting now, or should I wait?
Why does the joke only work with the word "sus"?
Because the whole routine is a physical translation of one word. According to Dictionary.com, "sus" is slang, mainly American, meaning suspicious, a clipping of the longer word that has been recorded since the late 1930s. It became a mainstream internet word much later, and the chant "that boy, that boy, that boy sus" turns it into a call-out: someone is acting shady and you are the one who noticed.
That is why every gesture points somewhere. The finger goes up and the head swings away, as if you are scanning the street for the suspect. The fist to the lens lands on "sus", and the deadpan stare afterwards is the verdict. If you strip the word out, the clip is just a person waving their hands around in a car. With the word, it reads as a tiny story with an accusation, a pause and a dismissal.
Is this the same trend as the 2020 standing dance?
Related, but not the same. A "that boy sus" dance was all over TikTok in July 2020, usually performed standing up in a bedroom, and it used the same chant. The October 2026 wave moved the joke into the front seat of a car, filmed with a phone held low at arm's length, and added the fist-to-camera jab and the wave-off ending that the older versions did not centre on.
Which song is it, and why do we not name one?
Because nobody has confirmed one. The 2020 posts and many October 2026 remakes carry a TikTok sound label that reads "hatchback" by cynthia ☆, but the car-seat posts mostly use re-uploaded "original sound" versions of the same short chant. A sound label on an app is the uploader's choice, not proof of the underlying track, so we do not claim a title or an artist.
For you this matters in one way. The videosi file already has the trend sound on it, so you do not have to hunt for it, and you do not have to guess which upload is the right one. If you prefer a different sound when you post, swap it from your app's own library. If you want to be careful about rights, treat the sound like any other music in a social post and keep to what the platform offers inside its own editor.
What does each gesture in the routine actually say?
There are six beats over nine seconds, and each one has a job. Knowing the job helps you judge whether a result is good.
- Point and turn away, twice. Finger up, head to one side, then the other. It sets up the search for the suspect and should feel slightly exaggerated.
- Fist to the lens. The head snaps back and the fist jabs at the camera on "sus". This is the peak of the clip, and the one frame most people screenshot.
- Deadpan stare. Nothing moves. The comedy is the contrast with the energetic start.
- Wink and temple. A hand goes to the temple with a wink, as if you are tapping your head: I know what I saw.
- Ear cup. A hand cupped behind the ear, listening for an excuse.
- Wave-off. Eyes closed, head tilted, a loose flick that dismisses the whole matter.
When you review a render, check the order first. A version where the stare comes before the fist, or where the wave-off is missing, has lost the logic of the joke, even if the face looks great. That is why the motion is taken from a reference clip rather than typed out: the order and timing are the hard part.
Why does the mouth not match the words exactly?
The clip comes with the trend sound, but the model animates the mouth loosely with the expressions rather than syncing every syllable. In this routine that is mostly fine, because the original character is not really singing either: the mouth hangs slightly open during the stare and moves with the expressions in the rest.
Should I pick "Just you" or the caricature look?
There are 2 looks, and they serve different goals. "Just you" keeps your face shape, build and hair exactly as in the photo, and is the one to pick when likeness is the point, for example a clip for friends who know you well. The "AI caricature" look is the one the viral accounts use: bigger ears, wider eyes, slightly oversized front teeth and a longer neck, while your hair, skin tone and outfit stay recognisable and the skin and light stay real rather than cartoon.
The caricature is funnier when the viewer knows the original, and it fits the trend because that is how the first big clips looked. It is also the riskier choice if the audience is not in on it. Exaggerated features can read as mockery when the person in the video did not choose them, so use it on yourself or on someone who is clearly enjoying the joke.
What makes a photo work for a hand-heavy clip?
The routine looks straight into the lens for most of its length, so a front-facing photo with both eyes open and your head and shoulders in soft, even light gives the best likeness. Whatever you wear in the photo is what you wear in the car, so a jacket or sweater that looks good on its own is a better choice than a plain T-shirt.
Avoid sunglasses, masks and large hats. The wink and the deadpan stare are the heart of the joke, and anything that hides your eyes hides them. Clear glasses are fine. A dark shot or a side angle is the other common cause of a result that does not look like you, because the model has to guess the missing side of the face.
Why do hands in front of the face cause problems?
Because this is the hardest kind of motion for any video model. A hand crossing the face hides features for a moment, and when the face reappears the model has to rebuild it. The result can be a slightly different nose, a melted finger or, rarely, an extra hand. Likeness drifts most during the middle of the clip, roughly the seconds when the fist, the temple wink and the ear cup happen.
This is also why typed prompts alone tend to fail on this trend. Without a motion reference, general video models wave their hands at random and lose the timing of the fist. The videosi version is built on Seedance with the reference and a structured prompt included, and we do not publish the full prompt, because the order of the instructions matters more than the wording.
Is it fair to make this about a friend?
Only if they have said yes. videosi accepts photos of adults who have agreed to it, and does not accept celebrities, public figures or anyone under 18. Photos and results are deleted within 30 days.
The trend itself adds a reason for care. The clip literally calls someone suspicious, and the caricature look exaggerates features. Put those together with a friend who did not ask for either and the joke can land badly. Ask first, show them the result before posting, and let them veto it. A good test is whether they would happily share the clip themselves.
What does one try cost, and what if it fails?
A standard nine second video (Standard model, 720p) is 8 credits. Better models and higher resolutions cost more, and there is no free generation. Credit packs start at $6.99 for 10 credits and never expire, and plans start at $29.90 a month. If a render fails or is blocked, the credits go back to your balance.
The that boy that boy dance AI page shows the current examples next to the generator, which is a cheap way to see what a result looks like before you spend anything.
Is it worth posting now, or should I wait?
Now, if you want to be early. As of 28 October 2026 the car routine is in its first two weeks, and AI remakes are still few, mostly from a handful of caricature accounts, ordinary creators and a template account. That is a thin field for a clip with an earliest version that passed 20 million views.
