PROMPT TUTORIAL · UPDATED OCT 29, 2026

Toy car to real car train window AI prompt

Copy the prompt below into an image-to-video model, attach one photo of your car, and set a vertical 9:16 output. It describes the full sequence: a miniature on a train windowsill, a dark tunnel, and the same car driving out full-size.

What prompt creates the train window toy car to real car effect?

This is a general prompt written for image-to-video models that accept one reference image. It follows the same shot structure as the fixed TrainWindowAI scene, but it has not been tested on every platform.

One continuous 15-second photorealistic vertical (9:16) tracking shot. Use the car in the reference image as the only vehicle: keep its body colour, silhouette, wheel design and proportions identical, first as a miniature, then full size. No people, hands or reflections of people; no readable logos, text or number plates.
Keep a rear three-quarter view from the same side the whole time. No cuts, dissolves, fades or black screens.
0-3s: an approximately 12 cm long scale model of the car rests on the windowsill of a moving train; countryside moves backward outside. A concrete tunnel entrance approaches.
3-5s: the train enters the tunnel. Warm ceiling lamps and reflections; the window and miniature stay visible. The model's red tail lights switch on.
5-8s: the camera moves closer and lower toward the miniature's wheels. The window frame slides out past the edge of the shot without dissolving.
8-10s: the sill surface extends into asphalt. The same car is now full size and rolls forward inside the same tunnel.
10-12s: track the full-size car to the single tunnel exit and pass through it once.
12-15s: drive on an open dusk road under the sky, same rear-side angle, no second tunnel.
Audio: train rhythm becomes engine and tyre sound in the tunnel, then open-air wind and road noise. No music or speech.

How do you keep the same car before and after the tunnel?

• Use one clear daylight photo with the whole car, its wheels and paint visible, ideally a side or rear three-quarter view.
• Name the identity in the prompt (body colour, silhouette, wheel design, proportions) and say it applies to both the miniature and the full-size car.
• Keep one camera side and direction for the whole clip. Changing to a front angle gives the model a reason to redraw the car.
• Crop out other cars and people. Obscure readable number plates before uploading.

How can you adapt the prompt for Kling or other video models?

Use the model’s image-to-video mode with your car photo as the reference or first frame. If a model cannot generate 15 seconds, shorten the timeline proportionally (for example 10 seconds: 0–2, 2–4, 4–6, 6–8, 8–10) or make two clips, one ending in tunnel darkness and one starting from it, and join them in an editor. Replace “the car in the reference image” with the model’s own reference syntax if it has one. We have not run this prompt on Kling, Runway or Veo, so treat it as a starting point and adjust after the first result.

Why does the car shape change during the transition?

The model redraws every frame, and the change of scale is the hardest moment. It may simplify wheels, shift paint shades or invent details it cannot see in your photo. Hiding the change in tunnel darkness, keeping one camera angle and giving a complete photo all help, but no prompt guarantees an exact reproduction. Our examples page labels where this happened.

Should you use a DIY prompt or a one-click AI template?

Write your own prompt if you already use an AI video tool, want to change the scenery or camera, or plan to make many versions. Use a one-click template if you want one video without learning a model’s settings. TrainWindowAI uses its own fixed scene with Seedance 2.0 Mini, so you only upload a photo; see the cost and quality comparison.