"Cinemorpheque" is an immersive interactive installation inspired by classic Hong Kong cinema. The installation uses machine learning algorithms to recreate the four seasons from classic Hong Kong cinema, paying homage to the rich heritage of the city's film industry.

To achieve this, the technical workflow required translating historical cinematic source material into real-time interactive 3D assets, combining machine learning data processing with custom particle systems, hardware tracking integration, and virtual production rendering pipelines.

Custom Generative Adversarial Networks (GANs) were engineered and trained specifically for targeted image morphing, directly converting historical film archives and vintage celluloid stills into high-resolution, temporally coherent spatial textures. These trained visual datasets were integrated directly into Unreal Engine, where scripted Niagara particle systems utilized custom mesh arrangements to sample the visual data on a precise per-pixel basis. 

Virtual production shooting for the project took place inside a XR stage and LED volume—the core nDisplay cluster configurations were established within Unreal Engine to synchronise real-time rendering outputs across LED wall panels and camera tracking frustums, maintaining stable frame rates and correct perspective alignment during in-camera visual effects shooting. 

Throughout production, structured storyboard-to-runtime pipelines were maintained to translate director requirements into optimized real-time assets while coordinating directly with lighting technicians and stage builders to resolve physical-to-virtual constraints and manage on-set asset handoffs.

Following the completion of the virtual production shooting phase on the XR stage, the project shifted into a public-facing interactive art installation format designed for multi-venue exhibition tours. While the initial stage production focused on virtual production, the public exhibition configuration required a hardware-to-software interactive pipeline to translate physical user input into real-time visual changes.

This transition involved integrating sensor hardware into a designed enclosure to stream visitor movements directly into Unreal Engine, enabling users to control the generative seasonal simulations on site.