>>215381142
>>215381671
Humor aside it's a bit more than feeding lyrics in. You can manually edit MIDI stems and layer sound fonts on. There are features ("Persona") that preserve vocal elements (gender, timbre, cadence) as well as key leitmotifs (in MIDI form) of the exemplar song you pulled the persona from. There is also a feature that narrowa the pool of styles to pull templates from, to a playlist you curate. I can use this as a multigenerational selection mechanism to nudge the template design toward the style I want which is why a lot of the leitmotifs are the same between songs (once I've layered the rock opera persona on), so for example the arpeggios in Your Mom is the bassline from Dukat Did Nothing Wrong and so forth.
By feeding in success back to the pool and then generating personas based on the songs I like and then feeding that back into the pool, I'm reinforcing what I like similar to a Gen1 neural network. Then I create a genre song from the inspiration pool in whatever genre I want (disco, new jack swing, yacht rock, surf rock, other silly shit) and then I layer the rock opera persona onto it to normalize it with the rest of the tracks.
There's that episode of TNG where some crew members were being abducted by clickyclacky subspace aliens and they were in the holodeck trying to iteratively refine what table they remembered, and the method I just described is like that.
Covering one song (eg yacht rock) as another genre (rock opera) is dead easy since it's just MIDI stems and you can put whatever sound font you want on it. It's a harp arpeggio. It's a koto arpeggio. It's Swisgar and Toki shredding on grandpas guitars. In that sense the MIDI stem component is easy but the built in studio sucks and is Mario Composer tier.
