There are few modern IPs that conjure such immense and transportive imagery as the recently revived Dune franchise. When it was announced that there would also be a prequel to the main Dune story hitting HBO, a common question was what will the world look like?

When audiences sit down to watch Dune: Prophecy, they’re immersed in a world that feels impossibly vast, tactile, and lived in. For Julien Hery and his team at Rodeo FX, achieving that realism wasn’t just a matter of pushing existing visual effects pipelines to their limits. It meant exploring new technologies like Gaussian Splatting and asking how they could fit into large scale productions.

“I was aware of NeRFs,” Julien explained. “But I didn’t really pay much attention until I saw Gaussian splats. Once I had a chance to play with them, I quickly understood how they could be very interesting for us in visual effects.”

That moment of recognition came when Julien stumbled across CG Nomads’ GSOPs implementation for Houdini, the CG software at the center of his team’s pipeline. “That’s where my interest got piqued,” he said. “We were already using Houdini for FX, rendering, and lighting. Seeing splats running inside our primary tool made me think—maybe there’s a world where this could really work for us.”

The turning point came not in production, but in testing. “I first tried it on the Red Keep from House of the Dragon,” he recalled. “I ran our turntable renders through it, and suddenly I could look through the transparent windows. My colleagues and I were like, oh my god. This changes everything. We could review assets from angles we’d never been able to before.”

The team’s first real test came with the Imperial Palace, one of the largest and most complex assets in the show. “Scattering points across the whole landscape wasn’t possible,” Julien said. “Every single strand of grass was a CG model. If I had tried to scatter billions and billions of points, no machine could handle it.”

To accommodate this work, the team developed a hybrid workflow that combined synthetic cameras, drone captures, and deep images. “Deep images were crucial,” Julien explained. “Each pixel wasn’t just a color, it had depth. We could slice it, understand what’s in front and behind, and then accumulate those frames over time to create a legitimate point cloud.”

The process went something like this: render 900 images of the palace at 3K resolution, output the corresponding deep data, then convert those frames into a point cloud inside Houdini. 

From there, the data was packaged through COLMAP and trained in Postshot. Despite the size of the palace, Julien managed to keep the resulting reconstruction to around 18 million splats—small enough to run on a consumer friendly RTX 4090 GPU.

The real time sink was rendering the images in the first place. But once training began in Postshot, the turnaround was fast. “Because we weren’t solving cameras, it was synthetic, we already knew the exact paths—training was really quick. Within thirty minutes I could see something usable.” 

Julien is quick to emphasize that Gaussian Splatting isn’t yet a daily tool in production. “It’s still early days. I’d be lying if I said we relied on it for every shot. But I showed a demo to HBO, and we visualized the palace inside Apple’s Vision Pro using MetalSplatter. That blew people away. Suddenly you can imagine digital scouting—directors looking through virtual viewfinders, framing shots with a 24mm lens from inside the asset. The possibilities are huge.”

For now, he sees the most immediate applications in layouts and previs. “Showing a grayscale palace isn’t exciting. But if I can throw a camera into a splat and render something in real time that looks close to the final pixel—that changes the conversation. Directors can give meaningful feedback without waiting two days for a render.”

When asked about the ultimate potential of the technology, Julien doesn’t hesitate. “I have no doubt final pixels will be splats. The quality is already there. The main bottlenecks are color preservation, HDR support, and relighting. Once those are solved, it’s just a matter of time.” He sees parallels to how Unreal Engine crept into VFX pipelines over the last few years. “A couple of years ago, we would never have considered rendering final pixels in Unreal. Now we do. I think the same will be true of Gaussian Splatting. You can’t ignore it. It has to be looked at and tested.”

The journey also wasn’t without hiccups. Unreal Engine plugins at the time struggled with the sheer size of the asset, producing flickering and depth sorting issues at kilometer scales. Even Apple’s viewer occasionally needed reduced splat counts to remain stable. Compression formats like SPZ weren’t available yet, though Julien noted that the raw PLY files were surprisingly manageable in size. With more modern compression libraries like SOGs and .SPZ, in addition to more powerful VR apps like Spatial Fields, this bottlenecks can be mitigated.

Inside the studio, adoption still requires persuasion. Colleagues accustomed to long renders and perfect final frames can be skeptical. “It always takes convincing,” Julien admitted. “People have workflows built over ten or fifteen years. You can’t replace everything on day one. So we start small, show the obvious wins, and then build from there.”

And then comes the common question: how many triangles are in this? The truth, of course, is that there aren’t any triangles at all. That’s the point. Gaussian splats live outside the polygonal world that has defined computer graphics for decades. The palace isn’t made of mesh. It’s made of millions of overlapping Gaussians, blended into a lifelike 3D radiance field.

The early experiments on Dune: Prophecy may not have shifted the entire pipeline overnight, but they proved Gaussian Splatting could scale from demo to production asset. “We try to find the right project where we can test something new and push the needle forward,” Julien said. “On every project now, I’m asking myself—where can we use this tech, and how can we adapt our pipeline to push boundaries?”

The Imperial Palace is one answer to that question. For audiences, it’s a set piece that anchors the scale of the series. For Julien and his team, it’s a milestone in how far Gaussian Splatting has come in a short span of time. “You need to see it live to believe it,” he said with a grin. “It runs in real time on my laptop. People don’t realize how big a change that is until they see it themselves.”