Something interesting is happening in the visual landscape of major studio films: audiences are developing an eye for the uncanny valley of AI-generated imagery, and they don’t like what they see.
The Texture Problem
Human beings evolved over hundreds of thousands of years to read faces, materials, and physical environments with extraordinary precision. We can detect wrongness in a rendered face at the speed of instinct, before conscious recognition kicks in. This is why the uncanny valley exists — and why it matters for cinema.
AI-generated environments and characters in 2025 are dramatically better than five years ago. Many viewers cannot consciously identify them. But something in the biology still reacts. The unconscious signal reads: not real.
What the Data Shows
The top-reviewed films of the past three years — Oppenheimer, Poor Things, Dune: Part Two, The Zone of Interest — all share a commitment to practical cinematography, real environments, and limited or restrained digital intervention. The correlation isn’t accidental.
The Economics of Quality
Practical effects are expensive. They require physical construction, skilled craftspeople, and the kind of on-set problem-solving that AI cannot replicate. But they produce images with a material authenticity that digital environments still cannot match.
The directors who understand this — Nolan, Villeneuve, Lanthimos, Glazer — aren’t being Luddites. They’re making a quality investment that audiences, even unconsciously, reward.
The Question Going Forward
As AI visual tools improve further, the question isn’t whether they can produce photorealistic images — they already can. The question is whether they can produce images that feel emotionally true, that register in the gut the way a real location or a real face does.
That question remains unanswered. Until it is, the greatest filmmakers will keep building things.