Case study 01
A social-deduction murder mystery aboard a stylized Victorian train, designed for scrutiny, suspicion, and readable player behavior.
Final outputCreative + technical goal
Create a readable Victorian train for social deduction: validate carriage scale before detailed production, organize clue-rich compartments with practical and indirect light, and make baked-lighting review fast enough to sustain creative iteration.
Built in Unity for a client-specific custom platform, Victorian Murder Mystery placed players inside a stylized social-deduction experience. One participant was the hidden killer; the others combined environmental clues with observed behavior to identify them. The world therefore needed to support investigation and conversation, not atmosphere alone.
Dense period detail could obscure clues and player silhouettes, while the miniature camera made small scale errors immediately visible. Lighting had to create tension and guide attention without flattening the behavioral read. Slow baked-lighting turnaround also limited how often the art team could evaluate a scene.
Reference + composition
Research focused on luxury train corridors, lounge and bar carriages, enclosed cabins, period materials, and investigation games built around searching, conversation, and suspicion. Pencil thumbnails tested camera angle, movement through narrow carriages, player sightlines, clue zones, prop density, and focal-light placement. Rapid blockouts then validated carriage proportions and became shared scale references for the art team and vendors.
Game reference · corridor readability
Game reference · social carriage
Game reference · clue search
Period reference · material hierarchy
AI-generated · concept phaseReference → target → final
The palette organizes investigation. Dark foreground mass frames the miniature stage, controlled period midtones keep players and clues readable, warm practicals establish local search areas, and cooler fill separates depth without weakening the mystery.
Final frame · material hierarchy
Final frame · lighting hierarchyPortfolio color-direction analysis
Key light
#D89A4AMidtone
#9B8974Shadow
#344B50Material / local
#5B2C2EAtmosphere
#F3E8D2Accent
#E6BF6AMood + visual target
AI-generated · concept phase“Explore broadly. Translate selectively. Author the final decision in engine.”
My story intent and visual direction established the target first. Leonardo and Stable Diffusion then helped compare warm practical light, cool indirect separation, palette, and clue-readable focal hierarchy before detailed production. ChatGPT and Claude helped structure research, storyboard beats, review questions, and production documentation. These studies accelerated comparison; final authorship remained with the art team and the decisions made in engine.
Team + pipeline
I translated the visual target into asset priorities, scale references, lighting-review criteria, and clear acceptance notes for the art team and vendors. Remote reviews separated composition, material response, clue visibility, and bake readiness, making feedback easier to act on and keeping each scene aligned with the same social-readability goal.
Visual targets, proportion references, review priorities, and acceptance criteria give on-site or distributed artists and external vendors a reliable path from first submission to approved asset, regardless of team size.
Performance constraints were translated into visible art rules so the art team could protect the intended read while making efficient production decisions.
Engine + performance
Final engine execution
Final direction continued inside Unity. Warm practical fixtures created local pools of attention, while cooler window and ambient values separated rooms, players, and architectural depth. Baked indirect response unified the miniature set without removing the contrast needed for investigation and suspicion. Trim sheets and atlasing kept related period surfaces coherent.
Optimization
I treated bake cost as an art-production constraint. Fast review bakes were separated from final-quality bakes, lightmap resolution followed camera importance, and static geometry coverage was reviewed so background detail did not consume the same budget as focal areas. Progressive GPU Lightmapper review-bake time fell from more than six hours to about 45 minutes, while milestone bakes preserved soft indirect light and controlled contrast.
Outcome
Progressive GPU Lightmapper review bake
A readable train-bound social-deduction experience in which clue visibility, player behavior, period atmosphere, and faster lighting review support the same mystery.