The high altitude trampoline AI trend is a 10-second clip in which a person steps off a dam, cliff or skyscraper atrium, lands on a tiny trampoline hundreds of metres below and bounces all the way back to the railing they started from. It is not real; every clip is made by an AI video model, and on videosi (videosi.io) you can make yours from one photo of yourself, with no prompt.
This guide answers the follow-up questions people tend to ask next: why the joke works, why the physics cannot, which location to pick, how to shoot the photo and what to do before you post.
Key takeaways
- The trend is a 10-second AI clip: a jump off a huge drop, a bull's-eye landing on a tiny trampoline and a full rebound to the top.
- It is funny because of scale. The trampoline is a dot from above, and the drop is far beyond anything a mat could survive.
- Your photo is the only source for your face, hair, build and outfit, so a clear head-to-knees photo matters most.
- A standard 10-second video (Standard model, 720p) costs 9 credits; there is no free generation, and a failed render returns its credits.
- The clip looks like real stunt footage, so turn on the AI label when you post.
In this guide
- Why is the high altitude trampoline video funny?
- Could a trampoline ever catch a fall like that?
- Which of the five places should I jump from?
- What photo should I use for the jump video?
- Why does my face drift during the fall?
- Why does the clip end back on the railing?
- What sound should I add before posting?
- Do I need to label the video as AI?
- How is a CapCut or Dreamina template different from videosi?
- How much does one jump video cost, and how many tries will I need?
Why is the high altitude trampoline video funny?
The joke is scale. In the wide shots the jumper is a speck against a wall of concrete or rock, and in the top-down shot the trampoline is barely a dot with a ring of spectators around it. Your eye does the maths before your brain objects: that is a very long way down, and that is a very small target.
The second half of the joke is confidence. The jumper steps over the railing without a rope or a parachute, hits the exact centre, and the mat throws them straight back up. The clip then ends on a close-up of an out-of-breath grin, as if nothing unusual happened. The page for the trend breaks the ten seconds into five beats: the edge, the leap, the bull's-eye, the rebound and back on top.
Could a trampoline ever catch a fall like that?
No. A person who falls freely does not keep speeding up forever, but the speed they reach is far beyond what any mat could absorb. NASA's educational pages explain that gravity pulls a falling body down at about 9.8 m/s² near Earth's surface, while air resistance grows with the square of the speed until it equals the weight. After that the object falls at a constant speed, its terminal velocity.
For a skydiver lying face-down the figure is about 55 m/s, or roughly 200 km/h, according to the Wikipedia article on terminal velocity. As a rough illustration of my own, ignoring drag, a 300 metre drop would take close to 8 seconds and end near 77 m/s, so in practice the jumper would be at or near terminal speed when they arrived.
Nothing about a stretchy mat changes that. It would have to stop a body at that speed over a few metres, and then return the same energy upward without losing any, which is the part the clip quietly skips. The rebound back to the original height is the impossible bit, and it is also why viewers rewatch it. Never try it for real.
Which of the five places should I jump from?
The five places differ in mood more than in mechanics. The trampoline, the fall and the rebound are the same in all of them; only the setting changes.
Giant dam is the closest to the clips that started the trend, with a pale concrete wall and a crowd along the railing, and it gives the strongest sense of height. Skyscraper atrium suits a city or office audience, because the balconies full of spectators make the drop feel crowded and loud. Desert canyon has the best light at sunset and works well for a more cinematic post. Sea cave cliff gives a bright, blue and gold look. Helicopter is the odd one out: instead of landing on a railing, you catch the skid and hang from it on the way back up, which makes it the best choice for a punchline.
What photo should I use for the jump video?
The jumper is on screen head to toe for most of the clip, so a full-length or head-to-knees photo gives the model far less to invent. Face the camera, stand in even daylight and keep the background plain.
Sporty clothes read best because they match a stunt jumper and keep their shape in the fall; long coats and dresses flap and tend to change. Leave sunglasses and hats off, since the close-up at the end is the payoff and covered eyes make the likeness drift right then. If your hair is long, tie it back, because hair moves most in the fall and the bounce.
Photos upload as JPG, PNG, WEBP or HEIC up to 10 MB. Use an adult who has agreed to be in the video; videosi does not accept photos of celebrities, public figures or minors.
Why does my face drift during the fall?
For most of the fall, your face is only a few pixels wide. The model has to keep an identity it can barely see, and it fills the gap with a generic face. That is why drift is common in the middle of the clip and why the final close-up is where likeness is judged.
Two things help. First, the photo: a sharp, front-facing picture with your whole body gives the model the most to hold on to. Second, the order of priorities in the prompt: videosi puts the identity lock first and pins the close-up at the end, so the face, hair and outfit from your photo are the first thing the model reads. Hands on the railing and the backflip are where invented detail shows first, so look there if something seems off.
Likeness is never guaranteed. If the result is clearly not you, try a different photo rather than the same one again.
Why does the clip end back on the railing?
Because the loop is the point. Returning to the same edge turns a one-way fall into a closed story: jump, land, bounce, arrive.
It is also a practical rule for the model. Without it, generated jumpers tend to land beside the trampoline, drop into the water or never come back up. The videosi version keeps one jumper, one trampoline and one location, and the jumper lands on the mat and returns to the same edge. The one exception is the helicopter option, where the jumper catches the skid instead.
What sound should I add before posting?
The video you download is silent, so you choose the sound. A short, punchy audio that peaks around the landing works well, because the bounce is the moment viewers wait for. A rising sound for the fall and a hit on the trampoline contact lines up naturally with the beats at about 3 to 5 seconds.
Add the sound inside the app where you post, such as TikTok, Instagram or YouTube Shorts, so you use their licensed library. Choose 9:16 for those apps. I have not named a song here because the trend has no confirmed original sound.
Do I need to label the video as AI?
Yes, you should. The clip is designed to look like real drone and phone footage of a real stunt, which is exactly the type of content platforms ask creators to disclose. YouTube's help page says creators must disclose AI-generated or altered content that seems realistic, and that a realistic scene that did not actually occur counts; the details are on the YouTube disclosure page. The same page says disclosing does not by itself limit a video's audience or its eligibility to earn money.
Other apps have their own AI label settings, usually in the posting screen, so check the one you use. Turning the label on keeps the joke a joke, makes it harder for anyone to mistake the stunt for a real event, and lowers the risk of a takedown.
How is a CapCut or Dreamina template different from videosi?
CapCut and Dreamina templates are a good way to try the trend inside a single app, and many tutorials point to them. Templates change often, so what you find may differ from what a tutorial showed. videosi is independent and not affiliated with TikTok, CapCut or Dreamina.
The difference is mostly workflow. With videosi you do not open an editor or a template. You upload one photo, choose one of five places and a format, press Generate and download a 10-second silent MP4 with no watermark. The prompt, the motion reference and the identity rules are built in, so every video follows the same five beats.
If you like to edit, pick a template. If you want a quick, repeatable result on the web that you can then edit anywhere, the videosi page is built for that.
How much does one jump video cost, and how many tries will I need?
There is no free generation. A standard 10-second video (Standard model, 720p) costs 9 credits. Better models and higher resolutions cost more, and the price is shown before you generate. Credit packs start at $6.99 for 10 credits and never expire; a 50-credit pack is $29. Plans start at $29.90 a month for 100 credits.
Plan for more than one try. The landing and the rebound do not always line up, and the face may drift on one attempt and hold on the next. A failed, timed-out or blocked render returns its credits automatically, but a finished video that you simply dislike does not, so check the photo before you press Generate.