What's the state of AI generated music now? Posted on March 25, 2023 by Anonymous What's the state of AI generated music now?
Music just needs the human touch, you know.
What about this one from google?
That arcade track is fucking rad.
ikr? it's scary good
Rising synth I would jam all day
It isn't that good you pathetic retards
I feel like the music generation work just doesn't get public code releases. It's easily getting up to acceptable results now like . Isn't anyone going to let us play with it?
>that 8 bit song
fuck those are actually soul
lets see you make something better tard
>modern music heavily relies on software
An AI could easily create music as long as there is software for it. Music is just numbers and patterns.
Many many books have been written about how to create nice sounding music in accordance with patterns and rules.
Why are these pics always a generic Will Smith I-Robot styled robot doing some vague tech shit or having math in the background? Is this the creative power of proompters?
It’s just copying what everyone else does, rather than invent its’ own music
Just like Excision and his ghost producers
If it steals a good MIDI from some japanese eroge gamedev that only like 5 degenerates know about you'll never have a chance of knowing the theft.
It is already widely used for youtube vlog intros etc.
it will take a long time since i don't think you can count on companies like Google, Microsoft and Open ai to take you there but literal whos like Elevenlabs.
riffusion is good, get the automatic1111 plugin
there's a riffusion plugin for automatic1111? I have only used the site.
sorry forgot to link my github for you
like i said in the previous post i don't know if it still works cause i haven't tested it for months
just installed it today; works fine.
output is cooler than i figured it would be.
thanks for your link anon
i'm the guy who made the plugin, does it still work? there's 16 issues opened on it but i've been busy with other stuff so haven't even looked at the repo in a couple months
there could be massive improvements made to the sound generation i think, training a better vocoder like hifigan or waveRNN, but would need the original authors' datasets, griffin lim sounds like shit by default
i think where they could've made it better too is trained multiple versions aligned to an exact BPM with audio files cut perfectly, then we wouldn't need img2img alignment hacks
experience suffering while you still can, human civilization is bout to enter a whole new phase and you'll be sorry you chased happiness too early