If you want, i can post a version that swipes a masked bar across the view area to show the raw footage rendered out though maya vs the overlay footage. Took to long setting up, im going to mess around with a few other approaches.
Prob - i was getting some unwanted sludge metal when generating D&D and beer related songs. It seems to pick out 3 genera keywords and then composes lyrics based on any superfluous information - sometimes picking up on instruments and stuff. All in all its a good front end and sounds better then suno.ai (in my opinion). There is also a slider that dictates when in the song it generates from, which natively is at the 30% mark. This is good because they have an extend song option that allows you to add music data prior to the generated content or after (intro vs extend). This further is neat because you can change the generation keywords to influence what is generated (i have not played with that). It does pick up on the types of voice you want and the types of singing emotion (perky, bubbly, sad, cool, so on). Thats all i have to report on this model, if you find anything tell me, especially anything that i can deploy offline. The hugging face has a facebook model that does audio, but not to this quality. sad.
Also, https://studio.moises.ai/player2/8992c7a2-4ae4-4c2c-9e56-e47040489d1c/?context=spliter This splits tracks into vocal/drum/guitar/other, and there are a few other ones out there that do the same thing. This is how i isolated the drum on the above to drive my mask. So if you have FL studio or something (yarr me matey), then mixing and remixing and pitch shifting/beat matching shouldnt be as much of a headache. You could generate 10 songs of style XYZ at 120 BMP in the key of F# minor and then take vocals from song A and B and add in drums of D mixed with E, for example.
Log in
