Well yeah, because music is not a modality of the models involved at all. It's literally just an LLM with the timestamps of each line injected into its context,

Well yeah, because music is not a modality of the models involved at all. It's literally just an LLM with the timestamps of each line injected into its context, and then making an API request to image / video generators. I don't really understand what the point of this project even was; stitching together dumber models with bespoke glue logic like this takes us further away from general intelligence and confuses the public.