Research Question
Can an audio file contain enough musical information to guide AI stage lighting decisions that feel intentional, structured, and usable in a professional show context?
The Vision
What if a lighting designer's years of experience could be encoded into a model — and used to generate a full light show from a single audio file?
MLI, short for Michels Lighting Industries, is a private independent AI stage lighting research project exploring neural audio-to-light synthesis, audio-reactive lighting, and automated stage lighting for live show contexts. The goal: study whether a model can learn meaningful relationships between music and lighting design without reducing the craft to beat matching or visual effects.
Current work is built around real concert recordings and careful evaluation. The central challenge is data: professional lighting shows are context-dependent, inconsistently documented, and difficult to compare cleanly. That challenge is part of the research question.
This is not a finished product. This is research only for now, documented with methodological honesty over hype.
The System (for the PoC)
Music Analysis
Audio is analysed for musical structure: rhythm, energy, frequency ranges, transitions, and larger changes over time. This is the foundation for audio-reactive lighting research, but the exact feature strategy is intentionally not disclosed.
Learned Decision-Making
Models trained on real show material study how human-designed lighting relates to music. Current work focuses on timing, intensity, visual energy, and consistency across different musical contexts.
Translation to Console
A deterministic translation layer turns research outputs into controllable lighting actions and console-oriented commands. The goal is practical control, not just visual approximation.
Current Status
Training Data
Active
96 real concert recordings processed. Dataset quality, consistency, and coverage are active research constraints. Current focus: scaling the dataset without lowering signal quality.
Model
In Progress
Core models are trained on real show material. Cue timing shows measurable learning. Higher-level content generation, style consistency, and generalization remain active research questions.
Product
Planned
Independent research that may later inform tools for lighting designers. Any future interface will focus on connecting AI-assisted lighting research outputs to professional lighting workflows. Target: late 2026 to mid 2027.
Get In Touch
Research inquiries, collaboration, data partnerships.
If you work in professional lighting design, Veranstaltungstechnik, AI stage lighting, or live production and are interested in this private independent research project by Casimir Michel — or if you have access to lighting show recordings and would like to contribute to the research — reach out.
[ LinkedIn ]External link disclaimer: LinkedIn is operated by LinkedIn Ireland Unlimited Company. The Owner of the current page is not responsible for external content, data processing, tracking, availability, or changes on LinkedIn after leaving this website.