Generative AI in music production: What do we do with it?

[Innermost]

Hi all.

I have already shared some message on this forum where I share some news about my Obsidian Neural project.
I've noticed some skepticism/pushback. And to be honest I can understand your point of view. A musician who would like an instrument who would generate a full track would destroy his own pleasure to compose music.
Then... Today Generative AI is here.
Then... What do we do with it?
Do we ignore it and continue as before?
Do we use it and abandon our "traditional" way of composing music in our DAW with our VST, samples, etc?
Or do we use AI with moderation, not to replace our work, but to generate weird sounds, weird textures, etc. Any sounds which would be interesting to use?

I imagine that some of you would claim that I tell you that in order that you download my VST and take a subscription on my API.
Honestly, that's not my main motivation. All the code is full open source on GitHub. You can make your own server. And if someone here has a GPU and want to launch a server, give some API keys to his friend, it's ok. because all the code is open sourced. And that would be great.
yes I have an API solution, because when I initialy released the project in June, a lot of people told me it was too much difficult to install my plugin, etc. Even a journalist told me that while my project wouyld be hard to install he wouldn't write any article on it (spoiler: since it's easier he doesn't even answer me anymore).

I know a lot of musicians hate AI.
I know that's legitimate.
I am a musician too.
I play guitar since my 6 years old.
I know there are f**king risks about AI (I invite you to read the investigation I wrote here about AI drifts: https://dev.to/innermost_47/when-ai-con ... study-1h0g)

Then... What do we do?
And you?
What do you think?

[Zeisner]

AI can be a great tool if you design and train it properly like it's done in scientific fields. Music AIs don't belong in this category, they're nothing more than slop lootbox machines. It's interesting to toy around with them out of curiosity, but for professional work? Hell no. It takes forever to even get 40 % of what you're aiming at when it comes to composition and arrangement and in the end you have to replace all stems anyway because the sound is atrocious. It's a waste of time and money.

[Innermost]

AI can be a great tool if you design and train it properly like it's done in scientific fields. Music AIs don't belong in this category, they're nothing more than slop lootbox machines. It's interesting to toy around with them out of curiosity, but for professional work? Hell no. It takes forever to even get 40 % of what you're aiming at when it comes to composition and arrangement and in the end you have to replace all stems anyway because the sound is atrocious. It's a waste of time and money.

What kind of generative AI have you tried to have this opinion? No judgement, just curious about your experience in fact.

[Ou_Tis]

Suno's Warner deal should give them access to a ton of stems, which should mean we can finally generate just an individual part (without having to generate a whole song and then use stem separation). Here's hoping they'll have enough to "ethically" train on from the Warner deal for it to sound good.

[lobanov]

What about personally me... I can imgine myself using AI tools to get some inspiration and ideas (I don't mean "steal"), it's possible. But I can use music of other people for the same purpose. And I get more from direct interaction with instruments or recorded sounds.

May be, I could be interested in developing my ideas. But I still see it as a trial-end-error process which may be futile.

Do I need assistance in composing (we are talking about generative AI)? No. What I am in need of is instruction and education.

May be I just don't want to learn really new things.

I hardly can say more regarding your questions. But not all people are like me.

P. S. Thanks for your effort anyway. For your programming and your questions.

[Zeisner]

What kind of generative AI have you tried to have this opinion?

Mostly Suno and Udio, with a few others I can't remember anymore (I only used them briefly to check if the same artifacts could be reproduced with them as well - the answer is yes). I ran several tests and none of the results were satisfying by any means, from instrument synthesis to composition to mixing. But as a musician and audio engineer I'm not part of the target audience anyway. I was just curious about this new technology and wanted to give it a try.

It can be a useful tool though to generate musical background noise for paper & pen roleplaying sessions on the fly. You don't need sound quality in such a case and the absence of strong captivating themes turns out to be a benefit (Less distraction). I also see potential for ads because nobody listens to them anymore, the era of catchy jingles ended decades ago anyway. But for "main" music, no. Just no.

[zwaarfris]

Here's a concrete use case, since "what do we do with it" is exactly
the question we kept landing on.

My bandmates and I have been writing and recording our own
compositions for 30+ years. Recording got technically easy, but
mixing and mastering never did, at least not without a budget we
don't have. So we ended up using AI as the production and
performance layer: we feed it a raw recording, let it build the
arrangement and performance around our composition, and the result
genuinely floored us. It sounds like we hired a 500k studio and a
session band. But the compositions and the core ideas are still ours.

Is it really ours, then? Yes, and also no, and I'm not sure I can
fully resolve that. But the question I find actually useful for a
production workflow is: what is the AI replacing? For us it's
execution, not creativity. The composition, the musical decisions,
the vision for how the record should sound, those are ours. The AI
handles what an expensive studio and session players would otherwise
have done. Whether that settles the "what do we do with it" question
I don't know, but as a way of using these tools it's the one that
made sense to us.

[Zeisner]

A user on this forum had an interesting idea: An AI with access to a synth, trying to replicate a sound of your choice. This could lead to very interesting results. Unfortunately, nobody will program such an AI, everyone just wants to jump on the LLM hypetrain.