Recreating Simon Meyer's Miniature VW Bus Commercial | Comparing AI Tools: GPT-5 and Veo 3

Simon Meyer's miniature VW Bus commercial went viral for a specific technical reason: it uses tilt-shift photography and forced perspective to make a full-size scene look like a toy being pushed around by an oversized hand. This video asks whether that effect can be recreated using nothing but AI tools, and treats the answer as a genuine comparison rather than a foregone conclusion.

The test setup

The recreation uses only GPT-5, Topaz, and Veo 3, with no traditional camera trickery. Seven scenes are built from detailed prompts, each describing the same 1960 Volkswagen Kombi Bus in white and pink with silver fenders and German plates, being moved by an oversized hand: an African savanna with a startled giraffe, a stormy coastal cliffside, an underwater coral canyon, a snow-covered alpine pass, a glowing nighttime cityscape, an indoor shelf scene, and a Weimaraner running toward the bus at Burning Man. Every prompt specifically calls out the tilt-shift technique and the miniaturized depth of field that makes the whole scene read as toy-like.

What each tool was actually good at

The comparison surfaces a clear division of labor. GPT-5 held the Kombi Bus consistent in its details, color, fenders, plates, across wildly different scenes and environments, but its output needed Topaz for upscaling before it was usable at higher resolution. Veo 3 took the opposite tradeoff: it handled sound and physical motion convincingly, which matters for a commercial built around a hand physically pushing an object through a scene, but showed real limitations when judged against the standard of serious production work.

Key takeaways

• Recreating a specific viral photography technique with AI tools requires precise, repeated prompt language describing both the subject and the camera effect.
• GPT-5 kept the vehicle's appearance consistent across very different environments but needed Topaz for image upscaling.
• Veo 3's strength was sound and physics, not raw output quality for professional use.
• The full prompt set for every scene is published for anyone who wants to replicate or adapt the experiment.

Try it yourself

If you want to run this comparison yourself, the complete prompts for every scene are published, so you can test GPT-5 or Veo 3 against your own reference footage. This kind of tool-versus-tool testing is a recurring thread in Nik Bear Brown's work for Humanitarians AI, which regularly stress-tests new AI video models against real production challenges.