Radiant Images Combines 24 iPhones with Gaussian Splatting for Next-Gen Bullet Time

by Alexandra ThompsonApril 25^th, 2026

Subscribe to News - Cameras

At NAB 2026, Radiant Images showed off a 24-iPhone bullet time rig that ditches traditional frame-by-frame interpolation in favor of Gaussian splatting. The result is a fully reconstructed 3D scene the operator can fly through with a virtual camera, retime at will, change focal length mid-move, and even edit at the point-cloud level. We caught up with the team for a hands-on demo.

Bullet time has been part of the cinematic vocabulary since “The Matrix” introduced it to mainstream audiences in 1999, and the core idea (freezing a moment while the camera flies through space) has barely changed in the decades since. What has changed is the toolkit. Smartphones, frame-accurate sync, and AI-driven volumetric reconstruction now make it possible to rethink the technique from the ground up. At NAB 2026, our own Graham Ehlers Sheldon sat down with Elnar Mukhamediarov, Head of Production at Radiant Images, to walk through what may be one of the most interesting evolutions of bullet time we have seen in years. Radiant’s broader Meridian 4D Volume Stage, built around exactly this approach, just picked up the CineD Best-of-Show Award for Camera Control and Accessories at NAB 2026.

If you want a wider primer on bullet time and volumetric capture as storytelling tools, we covered the basics in an earlier deep dive, including the iPhone-based rig used on “28 Years Later.” Radiant’s NAB demo takes the same underlying philosophy and pushes it considerably further.

Why 24 iPhones?

The first thing that catches the eye is the rig itself, an array of 24 iPhones standing in for the traditional ring of cinema cameras. According to Elnar, the choice of phones is deliberate rather than budget-driven. Each iPhone is an untethered computer in its own right, which means the team can develop its own software for the device, access files directly off the phones, and avoid the cabling and sync infrastructure a traditional bullet-time array normally demands. The iPhones also bring multiple focal lengths in a single body, so the operators can effectively change lenses across the array without physically swapping glass.

From traditional bullet time to Gaussian splats

The bigger conceptual leap, however, is on the processing side. Traditional bullet time stitches together still frames from each camera and interpolates the gaps to smooth out the move. Radiant’s approach instead feeds the captured frames into a Gaussian splatting pipeline.

For anyone unfamiliar with the term, Elnar gave a concise on-camera definition: a Gaussian splat is a volumetric processing technique in which an algorithm first works out where each camera is in space, then builds a point cloud, and finally compares that point cloud to the original images thousands of times until it converges on something visually close to reality. The output is a full 3D reconstruction of the scene, into which a virtual camera can be placed and animated.

Crucially, that means the operator is not flying between captured viewpoints; they are flying through a continuous 3D space. As Elnar put it during the demo, the workflow does not interpolate between frames; it flies through 3D space. The motion comes out smooth by design, because the move never relies on perfect pixel-by-pixel alignment between physical cameras.

Beyond 24 frames

Once the scene exists as a 3D reconstruction, the count of physical cameras stops dictating the length of the move. With 24 phones in the array, a traditional bullet time setup would yield a 24-frame sequence; Radiant’s pipeline can stretch the same capture to whatever duration the shot needs. The virtual camera can also retime itself, change focal length mid-move, and shift its perspective in ways that would be impossible to plan around with a fixed array.

In the on-floor walkthrough, Elnar demonstrated this by setting a key frame at the leftmost camera position, flying to the last camera, then dropping an extra key frame in the middle that pushed in closer and switched to a wider lens for an exaggerated low-angle look at Sheldon’s foot. The kind of move that would require either a second array or a heavy compositing pass in a conventional bullet time workflow becomes a few key frames in software.

Editing in 3D space

Because the underlying representation is a point cloud, the toolset extends well past camera moves. Elnar showed Sheldon his own digital reconstruction, then highlighted his hand inside the 3D scene, isolated it as its own collection of points, and deleted it. Returning to the virtual camera view, the hand was simply gone from the scene, with no rotoscoping or 2D paint involved.

This kind of object-level manipulation is one of the practical advantages of Gaussian splat representations over baked 2D plates. Cleanup, replacement, or stylization of specific elements can happen in 3D before any rendering takes place, and the same logic extends to set extensions, removals, and integration with CG geometry.

Service today, product tomorrow

For the moment, Radiant Images is offering the workflow as a service rather than a product. Production companies interested in using the rig hire Radiant directly, which handles capture and processing in-house. Elnar confirmed that the team is working toward making the system available either as a sold or licensed product down the line, at which point external rigs would be supported as well. The pipeline is also camera-agnostic, so although the NAB demo used iPhones, the same processing approach can run on motion picture cameras or any other synchronized array.

That camera-agnostic design is what underpins Radiant’s award-winning Meridian 4D Volume Stage, which packs the rig, built-in lighting, batteries, and a fully wireless real-time processing pipeline into 18 airline-checked Pelican cases. According to CineD’s Best-of-Show coverage, it can deliver real-time bullet time and Gaussian splat output at up to 4K 60p in ProRes 422 HQ or ProRes RAW HQ, with open-gate support and wide-gamut HDR via Apple Log 2, working interchangeably with motion picture cameras and the iPhone 17 Pro. More information is available on the Radiant Images website.

Does this kind of Gaussian splat-driven workflow make a 3D-reconstructed bullet time something you would actually plan a shot around, or does it still feel like VFX territory? Don’t hesitate to let us know in the comments below!