The Need for Speed AI prompt, explained
Last updated:
Every clip on this site starts from a fixed prompt. There is no prompt box, because the useful part is not a magic sentence — it is the order of the instructions: which input owns the car, which owns the scene, and what the model must never add. Here is the structure we use, line by line.
The five rules
1. Name the hero car and give the photo authority. The first sentence has to say that the vehicle in the reference image is the hero, and that its body shape, make, model, paint, wheels, lights, trim and modifications are the source of truth. Without this, the model averages your car with whatever car it has seen most often.
2. Say what must stay identical across shots. A single clip still cuts between angles. “Consistent in every shot” is the line that stops the paint changing shade or the wheels changing design halfway through.
3. Describe one action, not five. A burnout is a burnout: revs, weight transfer, rear tyres spinning, smoke, a launch. Adding a drift and a chase to the same eight seconds gives the model too much to resolve and it does all three badly.
4. Ban text before it appears. Badges, number plates, captions and watermarks are the first things to turn to mush. Asking for none of them looks cleaner than asking for the right letters.
5. Close the frame to people. Our prompts ask for dark tinted windows and no visible people, which keeps the output away from faces — the part of an AI video that most often goes wrong, and the part our content policy is strictest about.
The burnout prompt, in full
Cinematic automotive commercial. Use the exact vehicle from Image 1 as the hero car: identical body shape, make, model, paint colour, wheels, lights, grille, trim and every visible modification, consistent in every shot. The car performs a burnout and a slow drift in an empty industrial lot at dusk, thick tyre smoke rolling across the ground, warm sodium lights behind it, camera low and close, slow push-in then a sharp pan as the car swings around. Realistic scale, weight and tyre physics. Dark tinted windows, no visible people. No badges, lettering, number plates, captions, subtitles or watermarks. No dialogue and no music.
The highway and night prompts follow the same skeleton. Only the scene paragraph changes: a coastal road at golden hour with a tracking camera, or a wet city street at night with neon reflections and a long drift around a corner.
| Scene | Where it happens | Camera |
|---|---|---|
| Street Burnout | Empty industrial lot at dusk, sodium lights | Low and close, slow push-in, sharp pan |
| Highway Pursuit | Coastal highway, golden hour | Tracking alongside, then low behind |
| Night Getaway | Rain-soaked city street, neon | Low and close, fast orbit |
Settings that matter
The model is Kling 3.0, run through Yir at 720p with audio off. Audio is off because generated sound clashes with whatever track you add in the posting app; silence gives you a clean clip to score yourself.
Duration is 8 or 14 seconds. Both are native to the model, so there is no padding or trimming. The output follows the aspect ratio of the first frame: upload a 16:9 photo and you get a 16:9 clip, which is why the site crops the photo before it is sent rather than stretching the video afterwards.
The reference is a single first frame. Kling also has a reference mode that takes a photo plus a reference video, but that mode needs two inputs and copies the source clip shot for shot. We deliberately do not use it: it makes the result a derivative of someone else’s footage, and it locks you to that clip’s timing.
Adapting it
For a vertical cut, upload a 9:16 photo and keep the car in the middle third of the frame. The prompt does not need to change, but adding “keep the whole car inside the centre third” reduces the chance of a wheel leaving the frame when the model pans.
For a motorbike, say “ridden by a rider in a dark helmet” — otherwise the model will animate a riderless bike. For a van or a truck, add “keep the ride height and wheel size from Image 1”, because models tend to lower tall vehicles to match a sedan.
Do not put a real number plate, a brand name or a celebrity in the prompt. It invites a trademark or likeness problem, and the model usually renders the text wrong anyway.
FAQ
Can I use this prompt in another video tool?
Yes. The structure — photo owns the car, prompt owns the scene, no text, no people — works in any image-to-video model. The wording may need small changes because different models use different names for the reference image.
Why does the site not let me edit the prompt?
Because the three scenes are the product. A free-text box adds a moderation surface and a support burden, and most edits make the car less consistent. Choosing the length is the only decision that changes the output.
Does the prompt guarantee my exact car?
No. It keeps the car close, but AI video can still change a wheel design or a badge. That is why we ask for a clean photo and why a failed render is refunded.
Your car’s chase scene is one photo away
Create my clips