LightingGen: A DMX Based Generation Method for Entertainment Stage Lighting

For the scenes of stage lighting design in entertainment concerts, this paper proposes a DMX-based generative model, LightingGen, to simulate the lighting design and programming process of human experts. Unlike previous emotion-based lighting design approaches, LightingGen employs a data-driven method from the creator's perspective, standardizing stage lighting control parameters through DMX signals to create a unified lighting feature representation called DULA. LightingGen incorporates a two-stage generative model: the first stage uses LightingCodec to generate latent representations of lighting features, and the second stage uses music as a condition to perform conditional sampling in the latent spaces of lighting. To validate our methods, A virtual stage environment was built to collect 2.62 hours of DMX lighting control sequences as dataset. Experimental results indicate that LightingGen can generate lighting control parameters that partially align with pop music. To our knowledge, LightingGen is the first work to use generative methods to model the relationship between pop music and stage lighting.