Inside the Splat Capture Behind MrBeast's 500M Subscriber Video

When MrBeast’s channel crossed 500 million subscribers earlier this month, the milestone video carried a production technique most viewers would never have clocked.

The live action sequence of the Floor is Lava was reconstructed as a dynamic Gaussian splat, so editors could fly virtual cameras through the scene long after the shoot wrapped. The studio behind it is Arcturus, a volumetric and dynamic Gaussian splatting outfit we spoke to back in late March. The lineage of dynamic volumetric splats in entertainment runs back through work like A$AP Rocky’s “Helicopter” music video, but a fast, occlusion heavy game show inside a plexiglass cube is a harder problem than a controlled music video set. What follows is an edited Q&A with the Arcturus team on how the shoot came together and where the reconstruction was pushed hardest.

Q: How did the shoot come about?

Arcturus: After Bilawal Sidhu posted about Arcturus, MrBeast’s team saw the content and reached out directly to learn more the next day.

Q: How many cameras and how much footage did you use?

Arcturus: 30 to 35 cameras, depending on the shot. Action was captured throughout the shoot day and processed based on an assembly from the editorial team at Beast Productions.

Q: How did you calibrate the cameras, so that when talent was onsite, everything was ready to go?

Arcturus: Our calibration toolset has several components. We calibrate per venue, and then adapt that calibration per scene and per frame; those techniques are trade secrets. Cameras constantly shaking or getting knocked around was a massive issue, along with the challenges of plexiglass between the cameras and the action, so the calibration had to account for that.

Q: Multiple people moving fast in one volume means constant occlusion and crossing. Where did that push the reconstruction hardest?

Arcturus: Lots of occlusions means less ability to rely on having multiple views of the action throughout the solve. Our low-camera-count techniques proved very valuable in these scenarios. Some of the contributing strategies include depth supervision, and multiview-consistent pruning and splitting.

Q: How did you deliver the reconstructions to the editors, and what did they actually receive — a viewer, a scene file, proxies?

Arcturus: The editors received our 3D editorial tool and compressed splat sequences, which they could use to scout camera positions. They used these to provide creative feedback, and we iterated until the editorial team was happy.

Q: How fast was the turnaround? Did the editors get same-day previews they could cut around, or was it a post-heavy handoff after the shoot wrapped?

Arcturus: For this project, some post calibration was necessary due to the extreme conditions of the camera placements and the numerous bumps and occlusions from props and contestants, so we iterated with the editors over calendar time to suit their production schedules.

Q: Were the camera moves planned on set, or discovered later once you handed over the reconstruction?

Arcturus: Discovered later — that’s one of the key benefits of this technology. In one key moment, there was no shot that captured the dramatic sub-second race to the finish of the “floor is lava” game; even though we weren’t able to reconstruct perfectly near the edge of the cube, the technology enabled viewers to see what happened.

Q: What’s a shot in the final video that simply couldn’t have existed with a traditional multi-cam setup, and what made it impossible the old way?

Arcturus: There are shots that start above the roofline of the building and come all the way down into the middle of the action, in places that would interfere with contestants in any traditional setup.

Q: How many shots survived to the final cut versus everything you captured?

Arcturus: We capture much more than we process most of the time, then cull down to takes that get processed and iterated, and a subset of those are chosen for final based on flow, continuity, length of cut, and so on. We worked on 8 selects, and 3 were included in the final cut.

Q: What’s the next capability you couldn’t pull off on this shoot that you want working by the next one?

Arcturus: If we ever did another plexiglass cube full of people, we would approach camera layout differently to maximize usable camera angles and clarity.

The world of dynamic 3D through YouTube is still in extreme infancy. Despite this, there have been a growing number of large creators dipping their toes into using 3D as a new medium. To learn more about Arcturus, please visit their website.