https://google-research.github.io/seanet/musiclm/examples/
Alright 
The Story Mode is crazy
3 Likes
Specifically about MusicLM :
EDIT : Adapted to make fit with this thread, after being moved.
For now i only know Google MusicLM which is creating music from text prompts. But we could you use this thread to AI generated music in general
1 Like
As for MusicLM, Google is holding it back at the moment, for the same reason that ChatGPTshould probably not be allowed in it’s current form. They are both trained on data that is not in the public domain, and so content created with them could be subject to copyright.
The solution would be to either, create versions trained on open source content, or to pay the content creators in some way when their content is used, and allowing them to opt-in, or opt-out, of this.
None of the samples i’ve heard so far include vocals and lyrics. Does that exist ? That would be killer.
“Pop song in the style of Taylor Swift, about three minutes long, with a great hook.”
apparently it does but as of now lyrics are “incomprehensible” . 
1 Like
Just like a lot of human made music !
2 Likes
It rendered my mumble rap prompt spot on!
1 Like
1 Like
Another option might be to only us the output from something like this for “ideas”, and then use your own ears and mind to create a final composition, that you are pretty well convinced doesn’t break other’s copyright. Then you’d have a chance to clear up any muddied or distorted sounds, and get some well sung lyrics going too.
Good post today from Robert Henke on this topic:
AI is perfect in re-creating existing things in a highly believable fashion, be it images, text or sound. Even the most outworld-ish visual creations and mashups are based on known visual tropes, this is what makes it so powerful. We see new things that are not really new. It is all HR Giger or Rembrandt or Studio Ghibli or Wim Wenders, etc…
( No, i am not dismissing AI as tool here, there are fantastic artists out there using AI, but their process is typically highly iterative and driven by a vision that goes beyond of 'how to make an image look like XYZ (which now ironically also can mean, like the stuff AI artist XYZ is doing
) )
However, if AI for audio would have been ready before the invention of UK dubstep, I doubt it would have been able to create a ‘Burial’ or ‘Kode 9’ sounding result. And even if one of the random seeded creations would have sounded a bit like it, it would not have been recognized as important, since the social interaction and resonance that is required for invention would not have been there.
AI in sound forces creative people to rethink their practice: If you cannot contribute anything that is highly personal and unique, the AI will do it better.
There is no point anymore in ‘how to produce a professional dance track’ tutorials.
The more i think of it, the more i find this comforting. Don’t compare, do what you really really want to do. Leave the rest to the AI.
12 Likes
This bit is excellent. Humans are still needed for creation and curation. AI can only look backwards.
1 Like
I’m comforted in the fact that AI still relies on gatekeeping, which no matter what will always be human in origin and disregarded easily if wanted.
I’d like to hear three AI engines jamming with each other. Well I wouldn’t LIKE to hear it, but I suppose I’m curious in a sideshow type of way.
2 Likes
it sucks.
that simple.
4 Likes
This is fun.
Matt Mullen the Tech Features Editor at Music Radar did a side by side comparison of Meta’s MusicGen against Google’s MusicLM, with 5 different text prompts. You might think of what you might produce for a fifteen second sample on these text prompts :
- ‘Optimistic pop music with upbeat synth lead’
- ‘Experimental IDM beats with edgy production’
- ‘Spooky shoegaze with angry drum solo’
- ‘Wistful country song with female vocals’
- ‘Clean 130bpm drum loop to use in music production’
Give 'em a listen and let us know what you think.
Does any of this inspire any ideas for you ?
1 Like
Personally I’m fascinated by AI generated music, and so called AI “art” in general, I approach it like some sort of alien artifact. Having said that, on the other side, most of the AI “music” sounds a blurred collage of already listened stuff, impersonal, shaded, mimicking the already done, not sure it can be defined art
Interesting article on the use of AI for the recreation of a yet to be disclosed new Beatles song. ( Maybe Now and Then ? )
They were able to isolate a John Lennon vocal, from a cassette demo using AI. The script apparently might have been something like : ‘That’s a voice, this is a guitar, lose the guitar.’"
Also as Ringo says they would never use AI to recreate Lennon’s voice.
( They won’t, but will others, and will they be allowed ? )
This also brings forth the potentials and issues of using AI for separating old recordings into individual parts, for remixing and such.
Where’s the AI Funk at, I wanna hear some Ohio Players???
