I’ve edited my first AI entity today.
My wife heard a cool song somewhere in a market, she wanted info on it.
Upon listen, it sounded like Joe Bonamassa.
I’ve used sound recognition on Deezer app and it gave me an old looking black guy.
I started add his music.
Oh, strange, he’s old but yet unknown, the blues is good (but it sounds like ripped off Bonamassa, completely), he’s old but has only released singles! And singles were all released this month!
First time I’ve wasted my time on fucking AI rip off!
Once identified as AI (should be subject to votes / peer review), could we ban them from MB? (my preferred option)
Or could we have a system that blocks any further edits once a new AI marker is set.
We could repurpose the data quality marker for that, as it has no efects any more, now.
Apparently, at the moment, Deezer is the only music streaming platform to be against AI.
But its detection tool cannot detect everything (proof is my 2026-04-10 artist was on Deezer, not tagged as AI), AI programs will always fight it.
As they are generating money off real artists, they are highly motivated, and not by the love of music.
This is why we should ban those AI, worse than thieves.
Although I deeply sympathize with your sentiments, I don’t think this is the correct way to handle this. Thanks to your post, I now saw the list of AI scams and I think it is really useful, if only to provide a resource to unwary consumers of music. So banning it from MB would be self-sabotage of a powerful weapon against AI slop.
As for the blocking of further edits: I think this should only be considered if somehow the adding of further information is threatening MB resources. Otherwise: the more information can be gathered about these AI scammers, the more useful it will be in battling the scourge.
Perhaps some sort of AI flag on the artist / release / whatever, and hide the entries by default similar to bootleg releases? Then when the entry is displayed, it can include a prominent notice that this is AI generated.
We could add the “artists” for marking them and not their content.
It just takes two clicks for them (1. generate / 2. upload), but more work and resources for us, and there are tens of thousands new AI tracks everyday.
Just having the artists with a big red banner of shame, would be enough for me, so that I know I don’t need to listen to these.
And it would protect our resources.
There’s already a userscript that blurs NSFW cover art based on tags (also see MBS-6686) something like that to hide releases is definitely possible.
A single AI flag might not be sufficient, there is music that’s fully performed by humans but the lyrics are AI generated, or music is human produced and written but the vocals are AI, or the cover art is AI generated but all the music is fully human-made. And what about AI-assisted stem separation for remixing?
There’s people who enjoy AI music and could want to scrobble or tag it
I have a very similar mindset, I wish we could just get rid of all AI generated content on the internet, but I also see the benefit of using MB as a way to identify (and therefore safely ignore / blacklist) AI content.
That script is so bad lol I really should update it one of these days. I think it was the very first script I wrote for MB
There’s not actually that many people who unironically “enjoy” listening to that stuff. It’s mostly consumed by those who prompted it in the first place, and those unfortunate enough to stumble upon it on accident (like jesus2099 did). This Adam Neely video goes into detail about it, you can skip to the relevant section (about 10 mins long) if the video seems too long, though I do recommend watching the whole thing.
Tag it as AI, add an AI artist type, etc. But gatekeeping/censoring here is wrong. If people don’t like it, they can avoid it, not listen to it, and not edit it. Also tags like “Fuck AI” are counter productive. It may be your opinion, but it’s more likely to be downvoted by others, or not be used. “AI”/”AI generated” will be much more likely to have other editors join in tagging making it easier for you to filter out.
As for the making money off real artists, take that fight to social media, streaming services, AI companies, and to your countries law makers, IMO.
(I personally don’t consume such music, but it’s not my place to tell others what they can listen to and not others place to tell me what I can)
I think AI music should be permitted on the database without passing judgement about its quality, tagged with an attribute. There are other algorithmically generated or low effort releases, which predate modern AI. Vulgar language should not be used to describe these releases to avoid entries appearing unprofessional.
Characteristic features of AI generated music are the following: Unedited tracks contain an impulse of noise or raised level at the start and end. The spectrum is quantized into bands approximately 1 kHz wide similar to MPEG Layer-2. It is very obvious in the L-R (side) channel. This results in warbled, fluttery stereo. Mellow sounds are usually accopanied by gritty noise that rises and falls with the main sound. The spetrum resembles MP3 encoding, possibly arising from the training set containing MP3 data. Spectrum extends only up to 20 kHz.
These features can be detected by ear or a basic spectrum analyzer in the absence of a dedicated tool.
I think this July 2025 rewording by @RogueScholar, of a non-human kind of joke from before AI music existed, should now be removed from this page, on the light of the new situation with AI scam.
Strong agree with the idea that we should store this data - and tag and display it appropriately - so we can help consumers make informed choices.
I don’t believe MetaBrainz claims to be ‘neutral’ or without impact regarding how we impact the broader music industry/cultural landscape.
It’s been an oft-stated goal at summits etc that we want to help artists and don’t appreciate the industry status-quo (where control and money is kept or taken from artists).
That said, I don’t think that means that we shouldn’t keep the data we serve as broad as possible. But in terms of how we flag or display the data, especially on our sites, I think putting artists first fits with the ethos.
Is it actually a joke? I’ve made (bad) music by manually writing C++ code to generate something that sounds vaguely like music. No AI at all, but probably qualifies as non-human. I’m glad that’s allowed on MB.
Sorry, I should have phrased that better. Came off unintentionally dismissive I think. Poorly phrased attempt to say: post it to socials, the average people person has no idea of such problems unless it’s shoved in front of their face(too much going on to keep track of every problem). Cancel your streaming sub and when doing so, say explain that is why you are leaving, money is the only thing a company like spotify understands. And contact your government representatives, they don’t know what the people want if no one is yelling at them about what is important to them(And I think we’ve all seen videos showing how little most know of technology in general).
And one of my main reasons for editing is to do the small part I can to help even the smallest artists in my niche by making sure they are added and all the external relationships possible are there so they show up properly in places that use MB data making it easier for people to find and follow them. So 100% on board with helping artists.
But, pandora’s box has been opened, and there is no going back to a world without this. Even real live human artists are going to be using it more and more. I’ve seen it in my niche(KPop idol from a old group released a single and she used AI to make her video, but was intentionally vague about what was and wasn’t AI)
It’s not the same as ordering AI to produce a sound with lyrics about this or that, and, by copying the music of this and that artists.
And even the lyrics are copied from real artist works, by the way.
.
