An onboarding animation you drag like a film, built from Figma frames, Kling video and SwiftUI. How Traveme works, what broke, and when it is worth it.
The best onboarding animation is one the user controls. In Traveme, a travel app onboarding I designed and built, the whole onboarding plays like one continuous film: swipe and the scene moves forward, swipe back and it plays in reverse, and you can drag through it with your finger like scrubbing a video. The suitcase opens, your things float out, and the headline curves around the suitcase, sharp at any size. The method: every screen starts as a Figma frame, an AI video model (Kling) only paints the motion between two of my frames, and all text and buttons stay real code on top. Below is how it works, the four things that went wrong, and when this kind of onboarding animation is worth it versus a simple Rive or SwiftUI transition.
The Traveme onboarding experiment is one of my AI design experiments: a travel app onboarding, built as a native iOS app with SwiftUI from Figma frames through Claude and the Figma MCP, with the motion made in Kling video. The experiment description is the shortest way to say it: an onboarding that "plays like one continuous film."
Most onboarding is a carousel. Three or four static cards, a row of dots, a skip button. People swipe through it as fast as they can. The idea behind Traveme was to make the onboarding itself the first moment people enjoy, the "wow moment" I always design first. If the first thing you touch feels alive, you are more curious about the rest.
I documented this process in detail in my X post "Is Figma dead? No.", and Traveme is the main example there. Here is the method.
Before anything goes to a video model, I clone the Figma frame and hide every title, button and label. That clean frame is the plate, and it is the only thing the model gets. Video models are bad at text: letters melt, wobble and get re-spelled. So the model paints the world, and the words go back on top as real code.

Even the curved headline that snakes around the suitcase is not from the video. Claude renders it along the exact curve from Figma and draws it on top of every frame, so it moves in and out with the motion.
Each screen is a plate. Screen 1 is the first frame of the clip, screen 2 is the last frame, and the model only makes the motion in between: suitcase closed, lid opens, your stuff floats out. Because the clip starts and ends on my exact frames, the animation lands on the design instead of somewhere near it.
I learned this one the hard way. Playing the forward clip backwards to go back a screen looked very weird. A lid closing in reverse does not move like a lid closing. So the back moves got their own clips. Reversed playback is only used for one case: when you change your mind in the middle of a swipe.
Separate clips stitched together feel like separate clips. My first version had transitions that were too long and not fluid. What fixed it: every handoff gets a short blend of a few frames into the next clip so there is no pop, the slow parts are flattened into one even move, and the clips are encoded so you can drag through them with your finger, forward and back. Now you scrub the whole onboarding like one film.
An onboarding animation like this is mostly fixing small things. Four problems taught me the most.
The 2% zoom. The first video model re-rendered every clip about 2% bigger and shifted it a few pixels. You do not see it watching the clip. You see it the moment the video starts on top of the still frame: the whole design jumps. My check now is simple. Put the first video frame over the Figma plate at 50% opacity. If the suitcase handle shows twice, the model moved my design.

Switching models. Claude wrote a small script that measured the zoom and offset and scaled every frame back. It worked, but the clips also had uneven pacing and a few dead frames at the start. So I switched to Kling 3.0, which held my exact frames with no drift.
The idle loop that breathed. At first I bounced the idle clip back and forth. It slowed down at both ends, so it looked like it stopped and started again. Idles are now 10-second forward loops where the first and last frame are the same, with the pacing evened out so it never settles.
The reverse that looked wrong. Covered above: backwards is its own animation, not a rewind.
This method is not the right answer for every app. Here is how I would choose.
| Approach | Best for | Trade-off |
|---|---|---|
| Static carousel | Utility apps where people want to get in fast | Easy to build, easy to skip, rarely remembered |
| Code-only motion (SwiftUI, CSS) | Simple transitions, UI elements moving between states | Light and sharp, but hard to show a whole scene |
| Rive or Lottie | Illustrated characters and icons that react to input | Needs an illustration style; see Rive vs Lottie |
| AI video film you scrub (Traveme) | Brand-led products where the first impression sells the product | Heavier files, more checking of every clip, text must stay in code |
For a dashboard or a finance tool, I would not do this. At ViewStats, the onboarding work that moved creator activation from 13% to more than 65% was about showing value first, not about cinematic motion. I wrote that up in onboarding UX best practices. For a travel, lifestyle or consumer brand app, where the feeling is part of the product, a film you control is a strong first moment.
When building was expensive, an onboarding animation like this would have been the first thing cut from the sprint: "ship it static, add the animation later." Now the motion is one of the cheapest parts of the product. The expensive part is the same as always: knowing exactly what each frame should be.
That is why Figma got more important in this workflow, not less. The Figma frame is the reference that both Claude and the video model get checked against. If you want to see the same idea applied to a website instead of an app, my scroll animation examples post breaks down a perfume page that plays and rewinds on scroll. And the whole process from idea to shipped app is in my AI design workflow. The prompts and source for Traveme are available on the Traveme resources page on Gumroad.
Figma is great for designing each screen and testing simple transitions with Smart Animate. For a film-like onboarding, I use Figma differently: each frame becomes a plate with the text hidden, and a video model animates between two exact frames. The text and buttons are added back in code.
It is good when it shows the value of the product or makes people want to continue, and when the user stays in control. It is bad when it blocks people who want to get in fast. Always let people swipe, go back, or skip.
Yes. Traveme is a SwiftUI app. Simple onboarding motion can be pure SwiftUI. For scenes like a suitcase opening, I combine SwiftUI with AI-generated video clips that the user scrubs, and keep every word as real SwiftUI text.
Because video models still melt, wobble and re-spell letters. Keeping text in code means it is sharp at any size, easy to translate and easy to change without regenerating the clip.
If you want an onboarding that people remember, designed in Figma and shipped as real code, you can hire a product designer who builds the motion too.
If you have a web or mobile app idea in mind or you need some advice about product design, feel free to get in touch ✉️