About the tool and our sample videos

Toy Car to Real Car AI is an independent project operated by ZHAO JUNSHI, the individual operator behind LoftNum Studio. Contact support@toycartorealcar.net for support, reports or corrections. We are not affiliated with vehicle manufacturers, toy brands, railway operators or AI model providers.

What is available now?

The public site includes six AI-generated sample videos, scene choices and a prompt guide. Customer photo generation and checkout are currently closed while we validate the complete delivery workflow and payment setup. Launch prices are shown on the pricing page; there is no subscription.

How we made the samples

We used GPT Image to create vehicle references inspired by real vehicle designs, then generated silent videos with Seedance 2.0 Fast through KIE using our reference frames and a camera-motion reference. We added audio after video generation. The videos shown are generated test results, not footage of real vehicles undergoing a physical transformation.

| Sample | Connected environment | Displayed clip | | --- | --- | --- | | Sports car | Train window, dark passage, racing circuit | About 15 seconds | | Off-road vehicle | Alpine scenery, tunnel, mountain road | About 14 seconds | | Short camper | Warm train cabin, tunnel, lakeside sunset | About 14 seconds | | Pickup | Western landscape, covered passage, canyon road | About 14 seconds | | Family car | Elevated city railway, covered passage, city street | About 14 seconds | | Small car | Coastal lights, tunnel, seaside road at night | About 14 seconds |

All displayed examples are landscape 16:9 at 1280 × 720. The sports clip is the earlier, longer test; the intended customer output is 14 seconds. Clip length does not promise a generation time. Watch the actual examples.

What we check, and what can go wrong

We inspect the car's direction, proportions, paint, cabin and scenery continuity, transition, and wheel movement. Some pickup, family-car and small-car samples still show a stopped-wheel impression in the final driving segment. A 14-second duration alone does not prevent a static ending. We disclose these defects rather than describe the examples as perfect results.

These tests do not yet prove that replacing both the toy and full-size car with an arbitrary customer photo will work reliably. That workflow must pass further checks before generation opens. A reference prompt expresses the intended result; it is not a guarantee of car identity or motion.

Sources and checks

This page describes our own retained generation tests and visual review, checked on 29 October 2026. For the effect's visual structure, see our trend explanation; for editable instructions, see the prompt guide. We do not claim to have invented the trend or know its first creator. The site does not publish user uploads to its example gallery.