What Real-Time Cinematic Production Actually Is
The industry term you keep running into is real-time cinematic production, and Tailor Soldier Spy is one of the most referenced case studies for how it works in practice. It is not a piece of software you download. It is a pipeline methodology built around Unreal Engine that treats game engine render loops as a substitute for expensive offline ray tracing. The short film released in 2016 by Framestone and Illuminating the Dark won a bunch of awards for proving you could render photorealistic narrative footage at interactive frame rates instead of spending weeks per frame in a render farm. I have set up three of these pipelines for client work. The first one taught me more than any tutorial ever did, mostly because everything broke in ways that tutorials do not warn you about.
Tailor Soldier Spy pipeline fundamentals
The core idea is straightforward. You build your environment, characters, and lighting inside Unreal Engine 4, use high-resolution photogrammetry scans where possible, drive camera animation through motion control or keyframe work, and then capture the output as a video feed. The "film" look comes from post-processing volumes, color grading passes, depth of field setups, and careful material tuning. It replaces Arnold or V-Ray rendering with the engine's Lumen or Screen Space Global Illumination depending on your engine version. Here is the part that trips people up. The quality ceiling is not the engine. It is your source assets and your willingness to adjust camera distances. Unreal looks incredible at two meters from a character, but it starts to show its hand at fifty meters unless you have hardsurface geometry that actually exists at that distance instead of fake prop placeholders. I learned that the hard way on a commercial shoot where we had to pull back for an establishing shot and the whole city block looked like a miniature diorama because the props were baked from low-poly scans.
Building the Pipeline Step by Step
Asset preparation
You need photogrammetry or very high-resolution photobashed sources. Polycam, RealityCapture, or Agisoft Metashape will get you mesh imports that UE reads cleanly. The catch is UV layout and texture resolution consistency. Do not mix 2K scans with 8K scans in the same scene without a strict renaming and folder convention, or you will spend three days chasing missing materials while your art director asks why the lighting looks wrong. For characters, I recommend retargeting from Mixamo or Rokoko source captures rather than trying to sculpt from scratch. Time is the actual currency here, not realism in every polygon.
Get the Full Details
Lighting and materials
Use HDRI from Poly Haven or PhotoSphere Archives as your base. Then add spot and directional lights to shape the mood. PBR materials should use scan-derived albedo, roughness, and normal maps. I use SpeedTree for vegetation because the built-in wind shaders save hours of manual animation work. A counter-intuitive tip that beginners miss: bake your GI before final pass, even if you plan to use Lumen. A lightmap bake at your final resolution smooths out micro-noise in reflections that looks terrible on screen. I discovered this when a client sent back the 4K deliverable with visible flicker in metallic surfaces that only showed up after compression. Baking fixed it in twenty minutes.
Camera and animation
Record camera moves with a Mo-Sys StarTracker system if you have budget, or use Blender geometry nodes to approximate motion control curves. Keyframe the camera in Unreal's Sequencer, set the render output to EXR sequences, and grade in DaVinci Resolve. Keep your timeline at the delivery framerate. Do not animate at 60fps and convert down later unless you know exactly what you are doing, because the motion interpolation will create artifacts that are nearly impossible to clean up in post. My standard setup runs like this: Environment scan or blockout, asset import with a strict naming convention, HDRI base lighting, material assignment using scan data, Sequencer camera animation, test render at full resolution on a single node to check for flicker and resolution issues, lightmap bake if needed, final multi-node render export as EXR, grading and compositing in Resolve, delivery.
This process takes roughly one week for a three-minute sequence with moderate asset complexity, assuming the team has all scans ready before Day 1. If you start scanning during production, plan for two weeks minimum because scan cleanup always takes longer than you expect.
Where this approach fails and what to do instead
Real-time cinematic production does not scale well for extreme close-up macro work, transparent glass surfaces with complex refraction, or shots that require physically accurate water simulation at feature-film resolution. The engine approximations show up in those cases, and no amount of post-processing hides it cleanly. When I hit those boundaries, I composite a hybrid approach. Render the problematic element in a traditional offline renderer and integrate it into the Unreal sequence. This adds complexity but saves the schedule. Alternatively, redesign the shot to keep the camera further away from the problematic geometry. That is almost always faster than trying to force a perfect render from the wrong tool. I also ran into a specific issue where Lumen global illumination was inconsistent between frames on a slow-dolly shot, creating a subtle brightness shift that looked like a defect in color grading. The workaround was to switch that particular camera shot to baked GI with a precomputed radiance transfer cache, even though it added six hours of bake time. The consistency across frames justified it immediately.
If you want to study the original Tailor Soldier Spy work, the breakdowns are available through the Unreal Engine community channels and the Framestone site. There is no single download link for a complete pipeline because every project tunes the settings differently. Start with the templates Unreal ships, then modify based on your deliverable requirements.
