Tracked Follow Camera

What it is

A drone-style auto-tracking camera, built in Unreal Engine, that follows its assigned character through the scene at a consistent angle. Two instances run simultaneously on every shoot — one bound to each of the show’s two characters, Potato and Frittata — autonomously framing them at the same composition take after take with no operator at the controls.

Built in Blueprints as Realtime Supervisor at Versatile Media (2023) for The Future Chicken Today Show, a real-time-recorded web series where camera output ships as final pixels and there’s no post-production camera fix to lean on. The tool ran on every shoot day across both seasons (~sixteen episodes between February 2023 and the end of Versatile’s involvement).

In short: a drone that follows its subject from a consistent angle, always frames them, runs autonomously while the show records.

The problem

In a real-time-recorded show, the camera output is the deliverable. There’s no post-production camera fix, so every camera on stage has to behave reliably for an entire take. A hand-operated virtual camera ties up an operator. A static camera covers wide angles but not dynamic close-ups. What the show needed was a camera that could autonomously deliver the always-framed, dynamic-close-up version of every moment.

The model was an autonomous follow drone: subject-locked, smoothly tracking, doing the work of a Steadicam operator without needing one.

What it does

Fixed-angle follow. The camera follows its assigned character through the environment, maintaining a consistent angle relative to the world. As the character walks across the scene, the camera trails along at the same offset and continues to frame them. Crucially, the camera does not track the character’s facing direction: when the character spins or rotates in place, the camera stays put at its consistent angle. The character can perform freely without the camera spinning around them.

Consistent framing. The camera’s look-at target stays on the same point of the character (head height) take after take, so the character sits in the same place in frame regardless of where they walk in the scene. Cuts between this camera’s coverage of two different moments read as continuous because the framing logic doesn’t drift.

One instance per character. Each character gets their own follow camera. Spawning a second instance and binding it to a second character is the entire setup for two-character coverage. They don’t share state, don’t compete for the same subject, don’t know about each other.

Autonomous during the take. No operator at the controls during recording. The camera does its job continuously from take start to take stop, hands-off.

Technical implementation

Built in Unreal Engine using Blueprints. Each instance binds to a single character’s head transform at scene-setup time and uses that transform to drive both its own position and its look-at target. Position is computed as a fixed offset in world space relative to the character’s location (not the character’s facing direction), so the camera trails the character through the environment at a consistent angle. Look-at points back at the head, framing the character at the same composition every frame. Damping on both the position and the look-at smooths the response so the motion reads as a real piece of tracking hardware rather than a mathematically perfect lock; the small lag and recovery curve are what give it the drone feel.

The first version of the tool actually tracked the character’s facing direction, repositioning the camera to stay in front of wherever they were looking. The behavior worked for walking shots but broke down the moment a character spun in place, with the camera whipping around with them in a way that was too disorienting to use in the final renders. I toned the logic back to a fixed world-space angle, which let the characters rotate and spin freely without the camera chasing their orientation.

In production

The tool ran on every Future Chicken Today Show shoot day during the studio’s engagement on the show, across both seasons Versatile Media delivered with WSS. Originally planned as a six-episode run, the engagement expanded to around sixteen episodes between February 2023 and the end of Versatile’s involvement. WSS has continued the IP independently since. The tool is reusable: the core behavior (bind to a character, follow at a fixed offset, keep them framed) generalizes to any real-time-recorded show that needs an autonomous subject-locked angle as part of its multi-camera coverage.