The Tesla Roadster AI video on Tesla AI Video follows a fixed 12-second timeline of seven beats, and an identity lock keeps the person, car or pet from your photo the same through all of them. Together with a motion reference clip, those two pieces are what make every render read as the Roadster ramp flip instead of a loose guess.
This post is the deep dive into how the timeline and the lock work. It does not print the prompt, and it does not teach you to rebuild it on the underlying model, because the point of the site is that you upload one photo, pick a ride and get the clip. If you are wondering why pasted prompts do not reproduce the result, that is covered on the prompt page. If you want the result rather than the theory, the generator is one tap away. Tesla AI Video is an independent fan-made tool, not affiliated with Tesla, SpaceX or Elon Musk, and the videos are AI parody, not real footage.
Key takeaways
- Each render is a bundle: your photo as the identity reference, a motion reference clip, the prompt text and fixed settings.
- An identity lock tells the model that the person, car or pet in your photo is the subject and must stay that subject for all 12 seconds.
- The seven-beat timeline fixes when the climb, the upside-down shot, the backflip at 6 seconds, the thrust and the landing at about 9 seconds happen.
- Wings, planes and badges are ruled out because the model drifts toward them by default when it sees "flying car".
- Faces can still be small while upside down and badges or wheels can still drift, so the prompt reduces failures but does not remove them.
What is the built-in Tesla Roadster AI prompt?
The built-in prompt is the text half of a four-part recipe that runs every time you tap generate on Tesla AI Video. The four parts are your photo, a motion reference clip, the prompt text and the settings: 12 seconds, the aspect ratio you chose and the quality tier. You only supply the photo and the ride. The rest is fixed by the site, so every customer starts from the same structure.
The text itself is long on purpose. It is not a one-line wish such as "my car does a flip". It names the subject, describes the set, assigns a job to each part of the 12 seconds and lists what must not appear. Parts of it change with the ride you pick: the bike, your car, the flying supercar, the pet on a skateboard, the traffic hop and the neon night each swap in their own subject, scene and closing shot. The ramp, the lighting and the timing stay the same.
What does the identity lock do?
The identity lock is the part of the brief that tells the model your photo is the subject, and that the subject must stay recognisably the same from the first second to the last. In the built-in text your upload is referred to as the reference image, and the person, couple, car or pet in it is the thing that rides the ramp. Everything else in the scene is invented around it.
Without that instruction, a video model treats a photo as loose inspiration. You may get someone with your hair colour and a different face, or a car in your colour with another body shape. The lock pushes the other way: keep the face, keep the paint, keep the markings, keep the clothing. It is the reason a hatchback flips as a hatchback and a spotted dog stays spotted.
The lock has a practical cost, and it is worth stating plainly. It works best when the photo gives the model something clear to hold on to. A sharp, front-facing picture in daylight is easy to lock to. Sunglasses, heavy filters, a crowd of faces or a dim night photo are not. If you upload a group shot, the model has to decide who the subject is, and faces from photos with three or more people are less reliable.
Why does the Roadster flip need a motion reference?
The Roadster flip needs a motion reference because words are a poor way to describe a camera move and a body in rotation. "The camera cranes up as the vehicle climbs the ramp and goes upside down" can be read a dozen ways. A reference clip shows one of them, and the model can follow it frame by frame.
In practice the motion reference supplies the things that make the clip read as the trend: the wide shot from behind, the crane up to look straight down the ramp, the way the ramp bends past vertical, the slow turn through the dark and the pull-back as the subject lands. The reference carries timing and path. The prompt text then says what is moving along that path, which is your subject.
This split is why the reference matters: the words say who is moving, the clip says how the shot moves. Tesla showed the stunt on October 15, 2026, at a SpaceX test site in McGregor, Texas: a curved ramp, upside down, a launch, a flip and a landing. Our clip is a parody of that shape, and the motion reference is how the shape survives.
What are the seven beats in the 12-second timeline?
The 12-second timeline is split into seven beats, each with a time window and one job. Giving every second a job is what stops the model from rushing the flip into the first three seconds or spending the whole clip on the approach. The beats below are the ones the site publishes for every ride.
| Time | Beat | What happens |
|---|---|---|
| 0.0 to 1.0 s | Roll in | Wide shot from behind: the ride rolls up to the foot of the tall curved ramp, red light strips on both edges. |
| 1.0 to 3.5 s | Climb | The camera cranes up to look straight down as the ride climbs, its shadow sweeping across the concrete. |
| 3.5 to 6.0 s | Upside down | The ramp bends past vertical and the ride keeps going until it hangs upside down under the curve. |
| 6.0 to 7.5 s | Flip | It leaves the top edge into the dark and turns one full backflip, small against the black sky. |
| 7.5 to 9.0 s | Thrust and land | A short burst of white thrust slows the drop and it lands upright in a rolling cloud of smoke. |
| 9.0 to 10.0 s | Roll out | Side view: the ride rolls out of the smoke along a glowing red line on the ground. |
| 10.0 to 12.0 s | Payoff | The closing shot of the ride you picked: a kiss on the bike, a car in the smoke, a pet looking at the camera. |
Two details in the table do most of the work. The flip is placed at 6 seconds and the landing completes by about 9 seconds, which leaves the last three seconds for the payoff. That is why a clip that "has no flip" or "flips twice" usually points to a render that went off the timeline, not to a mistake you made.
Why are wings, planes and badges banned?
Wings, planes and badges are banned because they are the three things a video model adds on its own when it hears "flying car", and each one breaks the clip in a different way. The prompt names them as things that must not appear, so the model has an explicit instruction to steer away from its default.
Wings and aircraft shapes come from the model's idea of what flies. Show it a vehicle leaving a ramp and it reaches for the nearest picture: a plane, a drone, a car with fold-out wings. That ruins the joke, because the trend is about a normal-looking vehicle doing something it should not. The real demo used cold-gas thrusters developed with SpaceX, so the built-in text asks for small white thrust jets under the body that fire only on the way down, and for nothing resembling a wing.
Badges are a different problem. Models cannot draw a logo reliably, so any brand mark on a bonnet or a wheel comes out as a smudge that changes from frame to frame. It also creates a legal and honesty issue: we do not want to imply a Tesla product. The built-in cars are original, unbadged designs, and the text asks for no logos and no readable plates on any vehicle. That keeps your own car looking like your own car instead of acquiring an invented emblem.
How does the prompt change from ride to ride?
The ramp, the lighting and the seven beats stay the same across all six rides, and three slots change: the subject, the scene and the closing shot. That is why each ride has its own page and why a photo that suits one ride may not suit another.
- Bike Ramp Flip: a plain black city bicycle ridden by the person or couple in your photo. Upload one or two people.
- Your Car Takes Flight: the car from your photo, driven by an unseen driver, with four small thrust jets that fire only on the way down.
- Drive the Flying Supercar: an original red two-seat supercar with the roof off, so your face stays in shot at the wheel.
- Pet Kickflip: a dog or cat on a plain black skateboard, one animal per photo.
- Hop the Traffic: your car rises on its jets and clears a row of five plain cars instead of turning a backflip.
- Neon Beach Night: the same flip in a pink-and-teal 1980s beach city at night.
You can see each of them on the rides page, and the trend page explains where the clip came from. The practical point is that the photo type is part of the prompt's design. A ride that expects a person needs a person, a car ride needs a car, and a pet ride needs one animal. Upload the wrong kind of photo and the identity lock has nothing sensible to lock to, which is the root of several failures listed in the fixes guide.
What can the prompt not fix?
The prompt cannot fix a poor photo, a tiny upside-down face, a wheel design that wobbles or the fact that every render is a new roll of the dice. Being honest about these limits matters, because the timeline and the lock reduce failures without removing them.
Here is where the limits sit. Faces are small while the subject is upside down, so the likeness is weakest in that stretch. Badges and wheels can drift, particularly on your own car. Renders vary: two runs from the same photo and the same ride can differ in details, and one may land the timeline more cleanly than the other. Plates are hidden in the render, but that is a feature of the prompt and not a promise about every frame.
The levers that remain are all outside the text. Change the photo for a sharper, front-facing, daylight picture. Pick another ride that suits the photo better. Pick a higher model tier: Pro and Ultra hold likeness better than Standard, at a higher credit cost. Or render again. A Standard render at 720p costs 10 credits, and across all tiers and resolutions a render costs between 5 and 93 credits. A render that fails or is blocked returns its credits automatically, but a render that finishes is charged, even if you do not like it. The fixes guide lists eight common problems and what to change.
That is the honest summary of the timeline and the lock: they move the odds in your favour, and the remaining odds are managed with your photo, your ride, your tier and another try. When you are ready, open the generator. If you want to share the result afterwards, the posting guide covers formats and labels.
Frequently asked questions
What does the identity lock do in the Roadster AI video?
It tells the model that the person, car or pet in your photo is the subject and must stay that subject for all 12 seconds, keeping face, paint, markings and clothing. Small upside-down faces can still drift, and a sharp front-facing photo gives the lock more to hold.
Why is the flip at exactly 6 seconds?
The timeline fixes the backflip at 6 seconds so the climb and the upside-down shot get their own time before it, and the thrust and landing finish by about 9 seconds. A fixed slot keeps the flip from happening too early, twice or not at all.
Does the prompt use the Tesla name or logo?
No. The built-in rides use original, unbadged vehicle designs and ask for no logos and no readable plates. Tesla AI Video is independent and not affiliated with Tesla, SpaceX or Elon Musk, and the clips are AI parody, not official footage.