Digital Media Research

This study introduces a tool that automates light management in Virtual Production (VP), enhancing communication between key roles and improving lighting techniques in Unreal Engine 5 (UE5) through deep learning (DL) algorithms. Preliminary interviews were conducted in the form of focus groups with 10 participants attending VP workshops, with varying levels of expertise in cinematography and VP. Results showed that the limitations of a VP set, specifically when it comes to lighting, created a steep learning curve on their role as VP Directors of Photography (DoP), as well as their communication with the Virtual Art Department (VAD). The proposed tool aims to automate a very basic procedure that would normally require extensive planning and collaboration between departments. Using DL algorithms, three models were created: One predicting the light position for an input image, one predicting the light color for an input image, and one predicting both. All three models were created and trained, yet the strong focus point was the color predictor model, as this was easier to test and verify. By creating a short film with this tool, the study evaluates the effectiveness of streamlining the VP process.