How to handle AI in MusicBrainz

Continuing the discussion from Implications of AI generated content, specifically in response to @iolomorganwg in this post:

starting a new thread, since this is more MusicBrainz related, and the original thread was a more general/MetaBrainz topic

I foresee two different ways artists can credit an AI generator (be it musical, visual, or textual). either crediting them directly, like you would an artist (Cover by Midjourney or cover art generated by DALL-E and heavily edited in post), or crediting them as a tool (lyrics by Bill Smith using ChatGPT). so far I’ve only seen the former, so I see no issue crediting them as artists

we’ve already got many other entries for software as artists, including many voice synthesizers and TTS engines, a few AI, and even one for Microsoft Windows… that last one has a credit for album artwork too, namely “album art by C:\Windows\System32\IMJP10.IME @ 569px wide”, which I added, but someone else had added the artist originally.


A lot of this seems questionable to me:

  • Microsoft Windows - MusicBrainz is credited with MIDI files found in 90s-era Windows releases. MIDI files can’t contain metadata AFAIK, so I think that the “Microsoft Windows” credit just comes from the titles of YouTube videos uploaded by some random person with no connection to Microsoft. There are already works for these MIDI files that are properly credited to the actual artists who created them, e.g. Work “Trip Through the Grand Canyon” - MusicBrainz.
  • GASHISOFT GXSCC - MusicBrainz has an annotation stating that it’s referring to MIDI-to-waveform software that’s run by humans. Per Edit #12332891 - MusicBrainz, it sounds like the program’s name was only used in the credits since that’s how the humans who ran the software decided to tag the files.
  • Google Assistant - MusicBrainz isn’t credited with any releases, recordings, or works. I think that the only reason it hasn’t been auto-deleted is because it has a few URL relationships. I don’t see the reason for this being in MB; as far as I can tell, it doesn’t have anything to do with music.
  • Google Translate - Relationships - MusicBrainz is similar – no releases, recordings, or works. It looks like you added a relationship to credit it with performing the voice of Kit - MusicBrainz (of type “person”, which I don’t completely understand – was this meant to be a character instead?). It sounds like Translate (or more realistically, one of Google’s TTS voices) is being used as an instrument or tool here rather than this actually being an artist.
  • Green House Mastering - MusicBrainz has a disambiguation stating that it’s a company that does mastering, so I’m not sure why this is included in a list of software.
  • Microsoft Paint - MusicBrainz did not gain sentience and release a punk cassette. It looks like this artist was created when this Discogs release was imported. As far as I can tell from the very limited data there, there’s a one-off group of humans who called themselves “Microsoft Paint”.

that is a fair point. many of those I didn’t create, so I can’t speak to all of those, but I did create a couple and used one other.

  • Microsoft Windows: I explained above, there’s a release which used the binary of a file (or something like that, not certain) for the album artwork. it’s credited to the file, and I figured that was the best way to enter that credit.
  • Google Translate: I think you’re right that Kit probably ought to be a character, I’d been thinking the same thing, just haven’t put in the edits yet… Google Translate provides the vocals on tracks of Kitcaliber. I’d also seen this kind of relationship done before with other artists voiced by TTS engines. @RandomMushroom128 also mentioned on the Discord they’ve heard Google Translate as a voice on tracks
  • Google Assistant: I added this entity because I’m almost certain that artists have used this voice like Google Translate above. I don’t know of any offhand tho… I added it so there’s a well-tagged, well-entered entity ahead of someone (possibly me) adding relationships to it. I may try and look for artists who use this voice in the next few days…

but we’re already getting a bit off the topic of how to handle AI in MusicBrainz… :melting_face:

also adding that there was a lot of discussion on the Unofficial Discord about this topic starting about here and here (not sure if those links will work if you don’t have a Discord account, but I wanted to link it either way)

lots of folks looking forward to a “Software” artist type (now with a ticket!), including @Jeluang, @iolomorganwg, @aerozol, @Victini, and myself, all with different approaches in the interim. there’s too much to summarize beyond that, and I don’t want to misrepresent opinions either… :melting_face:


The disambiguation for Microsoft Windows - MusicBrainz says “Uncredited composers of the built-in music within Windows OS”, which is referring to a group of humans. I suspect that that Bandcamp album’s cover image was created by a human taking a Japanese IME file from Windows and loading it as a bitmap. Crediting this artist with this cover image doesn’t seem correct.

I don’t understand why TTS engines should be treated differently from e.g. guitars. MusicBrainz doesn’t have an “Ibanez JEM” artist that’s credited with providing the voice of Steve Vai. I can understand an artist of type “character” for Siri or Alexa if they’re explicitly credited as performing in a duet, though.

LLMs and their audio or visual counterparts can create (or parrot) works with limited direction from humans, so I think that more of an argument could be made that they could be credited.

1 Like

I don’t quite know why, but TTS and other instruments don’t feel the same to me… they feel closer to Vocaloids than instruments… I dunno, maybe I’m suffering from internet brain… :melting_face:

also, just for clarity, what’s LLM stand for? I tried Googling and didn’t come up with anything helpful…

1 Like

Probably opening a can of worms here, but for those who hadn’t seen/heard it, here’s Colin Meloy performing a mediocre song whose lyrics and chord changes were generated by ChatGPT in the style of Colin Meloy. I don’t see “Sailor’s Song” listed in Colin Meloy - Recordings - MusicBrainz yet.

If this was to be added to MusicBrainz how would it be credited?

Probably to Ghostwriter977 or whatnot?

Sure that part is easy in this case but what about the vocals?

Is it rather “performed vocals” or “sampled vocals” He didn’t perform himself, so the latter I guess?
Although it reads like the “Samples from artist” relationship is meant to be a less specific one of “Samples material

Which one is more appropriate to use or am I thinking too hard about it?

I understand that Drake had nothing to do with this and this is just AI kinda “reverse engineering” Drake? In that case, I would not link Drake in any way.

1 Like

So, if “AI” covers (where your favorite artist covers your favorite songs through voice cloning) become more popular we should not add any relationships to the “original” artists whose voice was used but what about “performance of” relationships?

Are they “cover” performances of the work?

BTW, as expected the Drake “AI” release has been nuked:

Rather off-topic, link to existing topic contained

Relatedly I’m not sure how to enter those, for lack of knowing a better word, “video remixes” – if at all.

In the case of “Maestro Ziikos” there are millions of views, maybe Trump should’ve become a singer instead, lol:
Camila Cabello - Havana ( cover by Donald Trump ) - YouTube

I’ve already wondered about this last year:
Editing meme songs, and dealing with sample, mashup, and remix relationships - #8 by chaban


If the new recording only replaces the original recording’s vocal track, then I’d say that this should be entered as a remix. If the backing tracks were also “redone”, then a “cover” performance-of relationship seems appropriate to me. As to who to credit the cover to, I’d say the person who put it together, or [unknown] - MusicBrainz if they’re anonymous.

I’m sure there will be a gray area in the middle for people to argue over.

Yeah. If this becomes common enough, I guess we’ll need to have a new relationship like “Recording used generated voice in the style of Artist”.


this is now history

If anything comes of this, it sounds like her desire is to be credited directly:

would the title of such a song credit you as a featured artist, or how would you prefer it?

Sure - anything anyone wants. Im just curious what even happens and interested in being a Guinea pig.

(Aside: Nice of her to offer to split the royalties when (IFAIK, IANAL) it’s an unsettled legal question whether she’s entitled to any royalties from a song that uses a model trained on her voice. Also, she followed up her request for 50% royalties by saying that she likes the idea of “killing copyright”. :roll_eyes:)

1 Like

This seems to be getting out of hand:

Without any kind of guidelines adding “AI” counterparts for artists seems a bad idea to me.
Wouldn’t actual characters like SpongeBob need to be split it off into another artist(s) too?

Are these the “real” characters or “AI” artists?

If this was to be added I’d expect it to be credited to Playboi Carti with status bootleg:

I’ve been adding a few of those voice cloned recordings but so far resisted adding novelty artists. See this one for example which had its lyrics written by ChatGPT, video made by another AI tool (maybe Midjourney or Kaiber?) and the voice transformed:


Here’s an RYM discussion about this:

It sounds like they’re taking a “wait and see” approach, but they may be less affected due to policies banning music that’s impersonating artists (I don’t think MB has anything similar, but I could be wrong) and apparently also not including singles that are only available on free streaming services like YouTube.

Regarding the Grimes thing, it sounds like Elf.Tech (which I guess is the voiceprint software) already has a policy on how songs should be credited:

Featuring GrimesAI or adding GrimesAI as a main artist, secondary or featured artist is the approved method of crediting Grimes when using the GrimesAI-1 voiceprint.

From the same page, it sounds like they want to gatekeep which songs are released. Good luck with that!


This release just got added. Don’t even know what to think of it, so I’m just gonna abstain.

1 Like

Admit I don’t follow style threads closely so if this is answered (not sure how to best search) apologize.

On a Korean digital distribution site (Bugs) I came across this release:

Title is : " 춘향연가" and the artist credit is to “zipma”

Couldn’t find anything online about this “zipma”. The Bugs site has this comment:
“A song created with AI by excerpting part of poet Jeon Bong-geon’s long poem ‘Chunhyang Love Story’.”

I assume this sort of AI generated music will be a “thing” (This release shows up on Apple)

How to handle? Maybe like vocaloid (lately have seen references to Hatsune Miku in general insterest sites)?

scott s.