🤯🤯🤯 Music generated by finetuning Stable Diffusion on spectograms then generating any song possible

I could see LO FI be the first genre to be completely taken over by AI beats (LO FAI?), it's relatively simple, predictable and barely any vocals, could make a 24/7 livestream of perpetually generated LO FAI:

Driving Music by Solar Heavy

here is a colab notebook for anybody that wants to play with Riffusion: if you want a custom deployment for yourself, run it on Banana 🍌 with this:

Thank you @levelsio! My friend and I built Riffusion over the past few months and I was literally going to reach out to you directly to get your thoughts today--then someone found it and posted to hackernews this morning 😅

Music industry is getting hammered with terrible AI music right now. This is interesting

What u mean?

Levels from Nomad guru to AI guru 😀

I am just the messenger 😂

The artificial vocalization is amazing. AI creating new languages. eminem style anger rap :

OMG!! 🤯🤯 I wonder if finetuning GPT-3 for MIDI would work

I had read about using computer vision to recognize audio waves years ago and completely forgot about it. The idea is so simple yet so effective. Awesome use case.

as resolutions improve, quality of spectograph-generated music will too. but in the meantime, can plug this into something like NU-Wave or another audio upsampling model and improve outputs.

This already worked pretty well some time ago with stuff like SinGAN to perform "AI resampling". I know of at least one recent work that uses spectrograms: Problem is lack of phase. I guess this project uses Griffin-Lim but Griffin-Lim is crap.

Music industry gonna sue it 😬

@jamespoulos Humanity might not be able to handle this kind of thing when it gets really good. Will mankind rise up and smash?

Wild

@djtoddterry peep this.

@acwangpython

It’s like the music version of a deffusion image with the hand coming out of the head or 3 hands. It’s like it kind of sounds like something but doesn’t make sense at the same time.

Such a clever way to generate music!

Dude, you're on fire with #ai rn.

…but can it do screamo?

@Cephii1 How many industries still need humans in the next 5-10 years?

Next up: AI talking head with AI music background ? 😉

@Cephii1 A buddy showed this to me about a few days ago. I was like...If you can see Ai and hear AI ..it's a whole other world.

Is it capable of like, making music clips that follow conventional measures, and sound "not shit"? This is utter arse.

What do you think @akirathedon ?

@drscotthawley how does this compare to what you all have been building at @harmonai_org? If fine tuning image methods can achieve results this good - I can’t wait for specialized audio models!

One of the most creative applications for SD I’ve seen

I love how much AI is going to change things in the next decade, its really exciting

AI generated music has no soul man...

Wait till it gets a soul

giliran musisi di senggol @fachrezamuslim @alhamdvlillove

Shitty. Doesn't recognise the text at all. Playing random things

It is uuuuugly as hell

Brilliant application and innovation... Opens up a world of possibilities!

It even loops! This is amazing!