Google remakes The Wizard of Oz with AI for the Sphere

What happens when you try to project an 86-year-old classic onto a 160,000-square-foot curved display? You can't just upscale the pixels—you basically have to recreate the world outside the camera's frame.

That is exactly what Google and Magnopus are doing for Sphere Entertainment, revamping the 1939 classic The Wizard of Oz for the Las Vegas Sphere.

When the immersive film debuts on August 28th, 2025, audiences will see a version that fills the venue's mind-boggling 16,000 by 16,000 pixel screen. And to get there, Google Cloud and DeepMind didn't just use off-the-shelf upscalers.

Putting Uncle Henry back in the room

The teams deployed fine-tuned Gemini models alongside DeepMind's Veo 2 video generator and Imagen 3 to sharpen resolution, widen environments, and digitally reconstruct missing characters.

In standard cinema, framing limitations mean scenes cut back and forth between people talking. But on a giant canvas that wraps around the audience, rapid cuts can be disorienting, so the team used AI to keep everyone in the frame at once.

According to the Wall Street Journal, an early scene where Dorothy talks to Aunt Em and Miss Gulch now visibly includes Uncle Henry. In the original cut, viewers knew he was in the room, but he was strictly off-camera. In the Sphere version, he is right there, framed inside a much wider view of the house.

Training AI on 80-year-old call sheets

To keep these generated elements authentic, Google didn't just feed the model old film reels.

Engineers combed through Warner Bros. archives, pulling original shooting scripts, behind-the-scenes photographs, concept art, and floor plans. Feeding these into Veo and Gemini allowed the models to learn historical costume details, set lighting, and even the camera focal lengths used back in 1939.

Those archival notes also taught the system how off-screen characters should act when brought into the frame, ensuring their movements fit the scene.

Ravi Rajamani, global head of generative AI engineering at Google Cloud, told the Wall Street Journal that AI has touched over 90% of the movie. Sphere CEO James Dolan hinted that physical sensory cues—like heat, wind, and scents—will accompany the screening.

Why it matters

This isn't just a remaster; it is a full-blown stress test for generative AI in Hollywood-scale live entertainment.

For audiences, it turns archival film preservation into a theme-park-grade spectacle. If you visit Las Vegas after August, it looks like a must-see demo of how modern hardware and video models can breathe practical new life into legacy media.

Google made a clever move here by anchoring generative models to real production archives. Instead of hallucinating random backgrounds, Gemini and Veo were grounded in historical sketches and focal length records. It showcases Google's enterprise toolchain solving an impossible visual challenge at an absurd 16K scale.

However, we think the creative implications are a double-edged sword. Altering over 90% of a timeless classic and inserting characters into scenes where directors originally chose to exclude them borders on rewriting film history. There is a fine line between sensory immersion and dismantling the deliberate framing that made cinema an art form in the first place.