The basketball dunk AI trend is a 10-second video in which a person who has no business dunking crosses up a defender and slams the ball through the rim. To make one, upload a single full-body photo to the basketball dunk AI page on videosi (videosi.io), pick a court and a format, and press Generate. You write no prompt.
Key takeaways
- The basketball dunk AI trend is a 10-second clip in which someone who cannot dunk crosses up a defender on a street court and slams the ball through the rim; the caption joke is "dunk it and I'll let you grab it".
- videosi makes it from one standing, full-body photo with no prompt, in 9:16 or 16:9, as a silent MP4 with no watermark.
- AI dunk videos fail by hand because of small jumps, split balls and blurry faces; a motion reference and an identity lock fix most of it.
- A standard 10-second video (Standard model, 720p) costs 9 credits; there is no free generation, and a failed render returns its credits.
- There are no jerseys, logos or real games, and you should turn on the AI label when you post.
In this guide
- What is the basketball dunk AI trend, and how do I make it?
- Why is the basketball dunk AI video so funny?
- Why is it usually a mum, a dad or a friend doing the dunk?
- Which photo should I choose for the basketball dunk AI video?
- Why do AI dunk videos often fail: split balls, tiny jumps, blurry faces?
- Which AI model makes the clip, and why does it matter?
- Should I pick the golden-hour court or the night court?
- How do I write a caption like "dunk it and I'll let you"?
- What sound should I use, and should it be vertical or horizontal?
- Can I make a real game, a team jersey or a famous player?
- Do I need to label the basketball dunk AI video as AI?
- Can I make it on my phone, and what does it cost?
What is the basketball dunk AI trend, and how do I make it?
Tutorials for the trend started appearing on TikTok in early June 2026, and the most-watched of them walk viewers through a template that turns one photo into the clip. Who started it is unconfirmed, and so is the song people use, so this guide names neither.
The clip is generated, not filmed. On the basketball dunk AI page you upload one standing, full-body photo, pick the golden-hour or the night court, choose 9:16 or 16:9 and press Generate. Rendering takes several minutes, you can close the tab, and the result waits in My creations. You write no prompt, because the movement is built in.
Why is the basketball dunk AI video so funny?
The joke is the gap between the person and the move. Everything in the clip is played straight: the low camera, the defender stepping up, the crossover, the crowd grabbing their heads. The only thing that is wrong is that the player is someone who plainly does not dunk.
Timing helps. The video is six beats over ten seconds, from the still pose to the crossover, the drive, the gather, the slam and the crowd rushing in. The slam lands at about the 7 second mark, so viewers have seen the build and are waiting for it. Then the camera tilts up and the person hangs on for a beat before the crowd reacts. That pause is where people laugh and replay.
Why is it usually a mum, a dad or a friend doing the dunk?
A famous dunker would make no joke at all. The humour needs somebody the viewer knows cannot do it: a mum in sandals, a dad in work boots, a friend with a backpack still on. That is also why the caption works. "Dunk it and I'll let you grab it" or "if you can't dunk you can't come over" only makes sense when the person being challenged obviously cannot.
Use an adult who has agreed to it. videosi does not accept photos of celebrities, public figures or minors, and users must be 18 or older. Photos and results are deleted within 30 days.
Which photo should I choose for the basketball dunk AI video?
Choose a standing, full-body photo facing the camera in even light, with your arms free. The first second of the video copies your pose, and the model takes your face, build, clothes and shoes only from that picture. A face-only crop leaves it to invent legs and shoes, and a seated photo makes the start awkward.
Leave hats and sunglasses off, because anything over the face makes the likeness drift once you are in the air and the camera tilts up. Everyday clothes beat a jersey, and a dark top against the bright court reads well on a phone screen. The clothes and shoes in your photo are the ones you dunk in.
Why do AI dunk videos often fail: split balls, tiny jumps, blurry faces?
Four things go wrong again and again when people try this by hand.
- The jump is a small hop. Without a motion reference, the model usually lifts you a few centimetres and lays the ball in, so the joke is gone.
- The ball splits or vanishes. It can turn into two balls, disappear during the crossover, or pass through the rim from below.
- The player changes. Video models like to dress you in a jersey and make you taller and leaner unless your identity is locked first.
- The face blurs in the air. It is the hardest thing to hold while the camera tilts, especially in the first few attempts.
The basketball dunk AI page works around the first three with a motion reference clip, an identity lock written first and rules such as one ball that goes through the rim from above and five fingers. It cannot promise a perfect face every time. A failed, timed-out or blocked render returns its credits automatically.
Which AI model makes the clip, and why does it matter?
videosi uses ByteDance's Seedance video model, with Seedance 2.0 Mini as the default and Fast, Pro (Seedance 2.0) and Ultra (Seedance 2.5) one click away. ByteDance describes Seedance 2.0 as a multimodal model that accepts text, image, audio and video inputs, with image and video references that give control over performance, lighting and camera movement.
That matters here because the dunk needs both a photo for who you are and a reference clip for what you do. A plain text-to-video model has neither, which is why hand-made attempts so often come out as the small hop described above.
Should I pick the golden-hour court or the night court?
There are two courts: a golden-hour neighbourhood street court and a night court under floodlights. Pick the golden hour for a warm, friendly clip that looks like a Saturday afternoon, which suits the family version. Pick the night court when you want a harder, more dramatic look, with the floodlights picking out the rise to the rim.
Think about your outfit as well. A dark top reads well on the bright golden-hour court, while a light top stands out under the floodlights. Whichever you choose, the defender wears plain clothes and the crowd is a line of onlookers.
How do I write a caption like "dunk it and I'll let you"?
The caption is a dare with a prize, and the clip is the answer. Keep the line short and end it on the prize, so the viewer reads it in the time the pose is still on screen. Typical forms are "dunk it and I'll let you grab it", "dunk it and I'll let you come over" and "if you can't dunk you can't come over".
Make it personal. Name the person in the video ("Mum, dunk it and I'll let you pick the film") and keep the stakes small and warm. The clip comes with no text on it, so you type the caption in the app where you post, and you can change it without regenerating anything.
What sound should I use, and should it be vertical or horizontal?
The video is a silent 10-second MP4, so the sound is your choice. Add a trending sound in the app where you post. Because the original song is unconfirmed, this guide does not name one. Pick something with a clear build and a drop, and line the drop up with the slam at about 7 seconds.
For TikTok, Reels and Shorts choose 9:16, because the clip is made to be watched on a phone. Choose 16:9 only if you will post it on a wide screen or place it inside a longer edit. The format is picked before you generate, so decide first.
Can I make a real game, a team jersey or a famous player?
No. The videosi basketball dunk AI page does not make arena or league games. There are no team jerseys, logos, scoreboards or captions, and the defender wears plain clothes. The outfit is always the one in your photo, and celebrities and public figures are not accepted. That keeps the clip a harmless street joke, not a fake highlight.
Do I need to label the basketball dunk AI video as AI?
Yes, turn on the AI-generated content label when you post to TikTok. The clip looks like real phone footage, and an AI video is allowed when it is labelled, so viewers know the dunk is not real. Do not use it to pass off a fake highlight as real or to misrepresent someone else.
Other platforms ask the same. On YouTube, creators set "AI use" to Yes under Attributes during upload when the content is realistic AI, such as showing a real person doing something they did not. The rules change, so check the settings in the app when you post.
Can I make it on my phone, and what does it cost?
Yes. videosi is a web tool, so a phone browser is enough, and the MP4 downloads with no watermark. You can close the page while it renders, because the result is saved in My creations.
It is not free, and there is no free generation or trial. A standard 10-second video (Standard model, 720p) is 9 credits. Credit packs start at $6.99 for 10 credits and never expire, and plans start at $29.90 a month. Better models and higher resolutions cost more, and a failed render returns its credits. If you only want one dunk of your dad, a single pack is enough.