Toy Car to Real Car AI Prompt: Kling, Seedance, Genjutsu

A working toy car to real car AI prompt has two jobs: keep your exact car through a change of scale, from a die-cast model on a train window ledge to the real car on the road, and hit the beats people know: tunnel, tail lights, night highway. Here is a prompt for each of Kling 4.0, Seedance and AI image models, plus the Higgsfield Genjutsu steps. Rather not write a prompt? The toy car to real car AI generator here needs one photo.

Each video prompt edits a reference clip. Use one you made, not the viral reel, which is someone else’s work: make a base clip once with a plain, badge-free car, then swap your car in.

The toy car to real car AI prompt for Kling 4.0

Kling 4.0’s edit mode keeps every frame and changes only what you name. Upload the base clip as video 1 and your car photo as image 1, and start with “Edit @video_1”, or Kling makes a new video instead of editing yours.

Edit @video_1. Replace the car, both the die-cast model on the train window ledge and the full-size car on the road, with the car in @image_1. The new car keeps the original car's place and size in the frame, its angle, its direction of travel and every move, frame by frame, and takes its body shape, proportions, roofline, paint colour, wheels and lights from @image_1; for any side the photo does not show, continue the same design. On the ledge it is a palm-sized die-cast scale model of that car with glossy metal paint; once it leaves the ledge it is the real car at full size. Rebuild the whole body from @image_1 instead of repainting the original car. Never mirror the car. Its wheels turn whenever it moves, and it is still driving in the last frame. Its badges, emblems, model lettering and number plate are exactly those of the car in @image_1; never add or invent a badge or lettering it does not have. Keep everything else unchanged: the train window and the ledge, the scenery, the tunnel, the road, the lighting, the camera cuts, the camera framing, the camera movement, the timing and the video duration.

Our generator uses a prompt of this shape; we will update this one after our current tests.

A Seedance 2.x version with a reference video

Add the base clip as @Video1 and your car photo as @Image1. The beats are timed for a 15-second clip; for 10 seconds, scale each time by two thirds. No reference clip? Delete the first two sentences.

Use @Video1 only for the camera work, the timing and the transformation: one continuous handheld phone shot from a seat in a moving train. Take nothing else from @Video1, not its car and not its colours. In every frame the car is the car in @Image1, with the same body, paint, wheels and lights, first as a small die-cast scale model, then at full size, seen from behind and to one side with its nose pointing the way the train travels.
0-3 s: the model stands still on the window ledge in low sunset light; scenery slides past outside.
3-7 s: the train enters a tunnel and the picture goes dark; tunnel lights streak past the glass.
7-10 s: the model's two tail lights switch on and glow red on the ledge; the camera starts to push forward.
10-11.5 s: the camera passes through the window, the ledge becomes the tunnel road and the model grows into the full-size car, which drives on with its wheels turning.
11.5-13 s: the car drives out of the tunnel into a flash of white light.
13-15 s: the real car drives along a night highway under tall street lights, still driving in the last frame.
Vertical 9:16, photorealistic, phone-camera look. Exactly one car, with the same badges and number plate as the car in @Image1 and none added. No people, no hands, no other text, no watermark.

Higgsfield Genjutsu: Object Swap, step by step

Genjutsu is Higgsfield’s AI video-to-video tool. Object Swap changes one chosen object and keeps the rest of the shot; Motion Transfer keeps only the motion. For your car:

1. Add your train window clip; Genjutsu takes reference videos of 4 to 30 seconds.
2. Choose Object Swap and add your car photo as the reference. It takes several reference images, so a second angle can help.
3. Say what to change: “Swap the car, both the model on the ledge and the car on the road, for the car in my photo. Keep everything else.”
4. Check the credit cost shown, run it, and compare both stages with your photo.

We have not run Genjutsu ourselves: these steps follow its page (read 2026-11-01) and October’s tutorials.

Start from a still: the model on the ledge

Some people make the opening frame first with an AI image model (Nano Banana, GPT Image or Seedream) and animate it once the model looks like their car. Attach your car photo and use:

Photo taken on a phone from a seat in a moving passenger train at sunset. On the window ledge stands a small die-cast scale model of the car in the reference photo, with the same body shape, paint colour, wheels and lights, seen from behind and to one side, its nose pointing along the ledge toward the front of the train. Low warm sun through the glass, the window frame along the right edge, blurred hills outside. The model is sharp and fills about a quarter of the frame, so it clearly reads as a toy. Vertical 9:16, photorealistic. One model car only, with the same badges and number plate as the car in the reference photo and none added; no people, no hands, no other text.

For a last frame, use this prompt, then give both frames to Kling 4.0 or Wan 3.0, as October tutorials did:

The same car at full size on a highway at night, seen from low behind and to one side at the same angle as the model on the ledge, tail lights on, wheels turning, tall street lights and wet asphalt. Vertical 9:16, photorealistic. One car, with the same badges and number plate as the car in the reference photo and none added; no people, no other text.

Common failures and how to fix them

Or skip the prompt

Here the scene, timing and swap are set: upload one car photo and get a 14-second 9:16 MP4 at 720p, usually 5 to 10 minutes later, for $9.90. See the pricing, what the clip looks like, or go to the generator.