Flying Car AI Video Team
A flying car AI video works on TikTok for a simple reason: the impossible move happens immediately. There is no setup, no talking head, no explanation. The car leaves the ground in the first second, and the viewer stays to see how it lands.
That advantage disappears if you generate a landscape clip and crop it. Vertical is not a crop; it is a composition. This guide covers the framing, the hook, the caption space, the loop and the export settings that keep the clip sharp after the platform re-encodes it.
本次消耗 30 积分 — Pro 月付 $39.90 每月可做约 6 条，年付 $319 可做约 80 条。
Start with the frame, not the effect
Choose 9:16 before you write the prompt. A vertical frame has three zones:
- Top third: sky, buildings, or caption space. This is where text will sit.
- Middle third: the car and the action. This is the only zone the viewer is guaranteed to see.
- Bottom third: road, ground, interface overlays. Keep important detail out of it; platform UI covers it.
A prompt written for landscape often puts the car in the centre and leaves the top and bottom empty. In vertical, that reads as a small subject in a tall empty frame. Fix it in the prompt: "the car rises from the lower third into the top half of the frame" is a vertical instruction, not a landscape one.
The first second is the whole video
TikTok decides whether to keep showing a clip based on early retention. A flying car clip has one reliable hook: motion that starts immediately.
Three hook patterns work consistently:
- The mid-action open. The clip begins with the car already airborne. No approach, no run-up. The viewer arrives at the impossible moment.
- The ground-to-air snap. The car is on the road, then it is not. The transition happens before the first second ends.
- The impossible stillness. The car hovers, perfectly still, above a normal street. The contrast is the hook.
What does not work: a slow push toward a parked car, a logo, a title card, or three seconds of empty road. Those are landscape habits, and they lose the vertical audience before the action starts.
Write the prompt for the crop
Here is a vertical prompt that uses the frame deliberately:
The car launches from a ramp in the lower third of the frame, flips once, and lands facing the camera while open sky fills the top half. 9:16 vertical, bright daylight, punchy contrast, room at the top for a caption.
Note the explicit composition: lower third, top half, caption space. The model responds to spatial language. "Centred" and "fills the frame" produce different results.
For a quieter hook:
A silver coupe hovers half a metre above a quiet suburban street, perfectly still, while a normal car drives past underneath. Static vertical shot, overcast morning, ordinary street details, no dramatic lighting.
The ordinary setting is the point. The viewer has seen a hundred neon renders; an impossible car on a normal street is more surprising.
Leave room for the caption
TikTok's interface covers the bottom of the frame and the right edge. Two practical rules:
- Keep the car's wheels above the bottom 20% of the frame.
- Do not put text in the right 15%, where the action buttons sit.
If you plan to add a caption, write the prompt with empty sky at the top. If you plan to add none, fill the top with buildings or clouds so the frame does not feel unfinished.
Loop the clip
Short videos are rewarded for rewatches, and a clean loop produces them. A flying car clip loops well when the final frame resembles the first:
- Landing loop: the car lands where it started, facing the same direction. The viewer can watch it twice without noticing the cut.
- Hover loop: a slow rotation that returns to its starting angle.
- Fly-by loop: the car exits one side and the next play begins with it entering from the other.
Trim the first and last half-second before exporting. Generated clips often have a settle at the start and a drift at the end; cutting those frames makes the loop feel intentional.
Export settings that survive re-encoding
TikTok re-encodes everything you upload. Start with enough bitrate that the second encode still looks clean:
- Resolution: 1080×1920. A 720×1280 upload is visibly soft after processing.
- Frame rate: 30 fps is enough. Higher rates do not improve a generated clip.
- Bitrate: export at 10–12 Mbps for 1080p vertical.
- Audio: add your track in the editor, not in the generator. The model returns silent video.
Do not upscale a 480p draft to 1080p for publishing. Generate the final at the resolution you need; upscaling adds softness without adding detail.
What to write in the caption
The caption has one job: give the viewer a reason to comment or share. Three formats that work:
- A question about the impossible: "Would you drive this to work or keep it for weekends?"
- A location prompt: "Which city should the next one fly over?"
- A transparent label: "AI concept, not real footage. Made from one photo."
The third is not a conversion killer. Audiences reward honesty about AI, and platforms increasingly require it. A clip that pretends to be real footage can be reported and removed.
Music and licensing
Use the platform's commercial library. It is licensed for the platform, which is where the clip will live. Do not pull a track from a random YouTube video and upload it to TikTok; the audio match system will mute the clip or route the revenue to the rights holder.
If you are cross-posting to a client's account, check whether the client's account has access to the same library. Music rights are per-platform and sometimes per-account type.
A repeatable posting workflow
- Pick one stunt prompt and one photo.
- Draft at 480p, 9:16, eight seconds.
- Fix the composition, then generate the final at 1080p.
- Trim the settle frames, add a licensed track, export at 10–12 Mbps.
- Write a caption with a question and an AI disclosure.
- Post, then check retention at the one-second and three-second marks. If viewers leave before the action, the hook is the problem, not the render.
That last step matters more than any prompt tweak. A beautiful clip with a slow first second will underperform a rougher clip that moves immediately.
Frequently asked questions
Do I need a different prompt for Reels and Shorts? The frame is the same 9:16, so the composition carries over. Caption placement differs slightly; leave margin at the bottom for all three.
Can I post the same clip on several platforms? Yes, but upload natively to each one rather than posting a TikTok link. Native uploads get better distribution.
Does the model add music? No. It returns silent video by design, so you control the licence.
How long should the clip be? Seven to twelve seconds is the sweet spot for a single stunt. Longer clips need a second beat, which usually means two renders edited together.
Open the studio, choose 9:16, and generate a draft. The vertical composition is the part worth practising.