How to Shoot a Knight vs Dragon Scene in UE5 Virtual Production

Why this clip went everywhere

In mid-2024 a short behind-the-scenes video of a knight fighting a dragon on an LED stage exploded across YouTube and social. The striking thing was not the final composite — it was that most of the shot was final pixels in camera, rendered live in Unreal Engine 5.

The split-screen comparison (real stage on one side, final frame on the other) made two things obvious: the lighting on the actors came mostly from the wall itself, and the dragon and environment were running in real time, not added in post.

This tutorial reverse-engineers the likely pipeline so you can reproduce the same class of shot without a full Hollywood volume budget.

Stage and wall configuration

The setup was almost certainly a front LED wall plus floor or side returns, driven by nDisplay at high refresh (120 Hz common for ICVFX to reduce flicker for the camera).

A camera tracking system (probably OptiTrack or similar) fed real-time position/rotation/focal length into UE5 so the parallax on the wall matched the physical camera move.

Key detail: the environment was built to be 'good enough' from the specific camera angles used. Not every surface had to be hero; only what the lens actually saw during the take.

Lighting and VFX on the day

The dragon and any fire/magic were likely a combination of high-quality mesh + Niagara systems running live. The wall provided the main key and fill, with a few supplemental LED panels for extra rim or eye light.

Emissive elements in the scene (glowing runes, hot coals, sky) contributed real bounced light onto the actors and practicals. This is why the plate already felt integrated.

For the dragon itself, many teams use a mix of a hero skeletal mesh for closeups and a lighter proxy or flipbook card for wider shots where fine animation detail is not readable anyway. AI-generated flipbooks can stand in during previz and even make it to final if the camera is moving fast.

Sequencer and performance on set

The camera move was likely a mix of handheld or dolly + Live Link driven Cine Camera in Sequencer. The dragon animation was probably triggered or scrubbed from the same sequence so timing was locked.

On set they would have run with a 'VP High' scalability preset: higher Niagara budgets and shadow quality than the final game, but still constrained enough to hold frame rate at wall resolution and refresh.

The split-screen BTS was almost certainly captured with two record outputs: one clean feed of the LED wall + talent, and one of the final rendered frame from the render node.

How to try this at smaller scale

You do not need a 10-metre wall. Even a 4K LED panel + floor return plus decent tracking can get you 80% of the look for hero shots and promos.

Use the same Niagara systems you plan to ship in the game (or generate new ones quickly with AI Flipbook Generator) so there is no visual disconnect between the cinematic and gameplay.

Profile early. The first time you put a dense particle dragon breath on a 120 Hz wall you will discover what actually costs frames. Cut the expensive modules or reduce spawn rate before the shoot day.

FAQ