Why a CGI Car Never Needed an AI Label – and What That Reveals About Article 50

Automotive advertising has run on renderings almost continuously for over a decade without it ever being treated as a transparency problem. Campaign images come from CAD data, sometimes before the first vehicle has even rolled off the production line – photorealistic, with precisely calculated light, with paint and material surfaces no real car would ever show in that exact perfection. Nobody has ever demanded a labeling requirement for that. Since August 2026, Article 50 of the AI Act, by contrast, requires that AI-generated images which could appear authentic be clearly identifiable. Two kinds of images, both potentially indistinguishable from a photograph – and yet the law treats them fundamentally differently.

The line Article 50 actually draws

The difference doesn't lie in how real an image looks, but in how it came to exist. Article 50 applies when an AI system generates or manipulates content – a classic CGI rendering, by contrast, comes from a person making every single decision explicitly inside 3D software: camera position, material values, light sources, reflection behavior. The result can look deceptively real, but the path to it stays fully traceable and attributable to a responsible person. That traceability, not how convincing the image looks, is the law's actual yardstick. I already covered here how narrow Article 50's scope actually is – this process boundary is the same reason a CGI car has never fallen inside it.

Why this boundary makes sense in practice

From my own experience – Gieske Studios went through the path from photography through CGI to generative AI itself – this distinction is less arbitrary than it first sounds. With a classic rendering, a CGI artist ultimately carries responsibility for every visible decision, because they made it deliberately. With a generated image, that same volume of decisions compresses into a prompt and a trained model whose exact weighting even the user doesn't fully know. That missing traceability, not photographic quality, is the plausible reason lawmakers draw the line here and not around traditional CGI.

Where the line starts to blur

This is exactly where it gets uncomfortable in our own production work: modern CGI pipelines haven't been purely manual for a while now. AI-assisted upscaling, generative texturing, AI-estimated lighting inside an otherwise classic 3D render are everyday practice today, including in our own work. At what point does a production that uses AI tools at several stages but is fundamentally directed by an artist become an "AI-generated" image under Article 50? The law draws a clean process boundary between human and AI system – real work in a studio like ours increasingly moves right along that boundary, not clearly on one side of it.

What this means for studios like ours

The practical consequence is documentation instead of gut feeling: recording, for every production, exactly which steps actually involved a generative AI system and which ones an artist decided is becoming the real compliance work – not how realistic the finished image looks. The same principle applies to the choice of tools itself: whoever knows exactly where in the process a human made the creative call can also prove it if it's ever questioned. Whoever doesn't document it leaves that reconstruction to hindsight, in a dispute, after the fact.

Conclusion

A CGI car never needed an AI label because the law doesn't ask how real an image looks, but who made the decisions behind it. That boundary holds up as long as human and machine stay cleanly separated – and that's exactly what's becoming the real task in modern, hybrid CGI pipelines: not delivering better images, but knowing precisely where in your own process the machine took over.