Brian AI Project.

[SMiLE Brian]

And what do you know, the legendary BBS historian is talking sh*t…

[rab2591]

And what do you know, the legendary BBS historian is talking sh*t…

I don't think the criticisms being lobbed at this are entirely accurate or fair. To quote Hey Jude from earlier, "It’s important with this stuff to try to remain level-headed, and also polite and not dismissive."

To say it sounds "nothing" like Brian is obviously excessive criticism. I liken the criticism to someone mocking a CGI explosion from a 1990s Hollywood movie "that looks nothing like an explosion" - when in reality the computer helped digitally generate an image of fire, smoke, and a shockwave. Is it a real explosion? No. Does it look perfect? No. But to claim it looks "nothing" like an explosion is to melodramatically ignore that a digital artist took real-life visual elements/cues and created a digital representation that is very similar to a real life explosion.

Ethics aside, I think this is absolutely stunning. Reminds me of the story of the creators of Pixar making a very rudimentary short version of Toy Story - it looked utterly creepy and disjointed, but the visionaries (including Steve Jobs) saw the potential and invested heavily...they didn't say "those look nothing like toys, this fails miserably" - they saw the potential and ended up making a multibillion dollar company that changed digital animation forever.

I'm not saying that seltaeb is going to become a billionaire using someone's AI software to recreate Brian's voice, but I am saying that to quickly/rudely dismiss/criticize this work is fairly pretentious and ignores the possibilities to come. I can see why someone wouldn't like what they're hearing with these tracks (for a couple reasons), but it's also another thing to be excessively dismissive.

[HeyJude]

Again, I have to reiterate that, based on what I've seen from a broad number of reactions to AI stuff (pics, video, art/"paintings", sound), people need to accept that, setting aside examples of people who have ulterior motives in praising or panning this stuff, even among folks making good faith attempts to take this stuff in, there is a BROAD range of opinions.

We're seeing this right now. Some folks are hearing this stuff and are stunned with how amazing it sounds. Others think it sounds awful. Truly. I've played these recent recordings for a few people, asked for a few reactions. They've mostly been *not* good, and these are people who know their s*it and are not just old cantankerous fans who don't like change/technology.

As with many polarized things these days, this doesn't bode well for the future of AI simply from the point of view of reacting to it and gauging its quality. And I've definitely seen with visual art a bit of the Dunning-Kruger effect in play, where non-artists make AI art, and to them it looks *just as amazing* as something a real artist would make. Just like some people hear autotune and think that doesn't sound completely unnatural as far as what a human voice should sound like.

Do these AI recordings sound like Brian Wilson? It's a bit like being "pretty pregnant"; you either are or you aren't. A lot of new tech these days is being sold on "just you wait, it's almost there", and in some cases with AI I think that's almost certainly true. I think ironically that AI-ing a singing voice is going to lag surprisingly far behind any visual medium (especially photo fakes) as far as what can actually "fool" people, or what sounds so close or just like something to the point where one can listen to it and, like, enjoy it and think (or pretend?) it's the real thing.

[send me a picture and i'll tell you]

To me, the AI vocals sound much more convincing than AI photos.

Imagine the possibilities. Using Brian's current voice to do a lead vocal for What a Fool Believes, and then slipping it into his jukebox...

Actually, I'd like to hear that.

[rab2591]

To me, the AI vocals sound much more convincing than AI photos.

Imagine the possibilities. Using Brian's current voice to do a lead vocal for What a Fool Believes, and then slipping it into his jukebox...

Actually, I'd like to hear that.

Yeah the possibilities seem endless. And like with anything, I'm sure there will be very bad implementations of it, but also very good implementations of it. It's kinda like some SMiLE mixes aren't great, unoriginal, but others really capture the magic...some Smile mixes may not be true to the history of the project, but they sound amazing. And like Smile mixes, there are so many variables that can make them good or bad.

Likewise I think there are so many variables with this new technology. And I also think that it opens up an entire new universe to explore and tinker with. These mixes are cracking the door open. Just imagine what we'll be hearing, good or bad, in the years to come.

[HeyJude]

It's interesting that just about every actual artist/creator really finds AI stuff to be pretty awful. It's easy to dismiss these people as all just s***ing their pants because they're going to be replaced.

But these are also people that know these mediums better than those who only consume it. I think that's worth something too. As are the ears of people who have been listening to these voices for 30/40/50/60 years.

Awhile back someone did a piece of AI art that was something like "Calvin and Hobbes if they were a Laika film", and some people looked at that and thought "holy s**t, that's amazing!", and then a bunch of people pointed out how *not* like that it looked. Actual artists (and just Calvin and Hobbes and Laika fans) were able to point out the 37 things that were totally off.

I played those Brian AI tracks to someone who told me "Wow, that *does* actually kinda sound like Curt Becher!"

I for one find the AI separation technology to be far more fascinating, with far more room to credibly potentially "fix" things or "right" things in a convincing way that can truly be enjoyed as a listening experience. Random example: Fix those two Beatles reunion tracks (which I actually love even as-is) so Lennon doesn't sound like he's singing from inside a toilet (Free As a Bird) or on helium (Real Love).

[rab2591]

I work in a field that is very much targeted by AI right now (graphics//video). But I heard a great commentary on it recently - that AI, like any other computerized tool in recent years, is a tool...it doesn't have to be the end of your job as long as you utilize the tool. It's much more complicated than that, but that's the simple version of it. I look at the graphic/art AI technology the same way I'm looking at this audio AI tech: it may seem rudimentary now, but wait for the tech to evolve.

Such technology could very well mean the end of my job in the future, but oddly enough doesn't make it any less fascinating for me.

I played the AI tracks for my wife and she was also blown away that it was AI...yeah it doesn't sound perfect and there are clear flubs/imperfections, but to think that this is the first iteration of the technology that has Brian sing on songs he never sang on. It's quite a marvel.

[pixletwin]

These are amazing. Especially Guess I'm Dumb. I would love to hear A Day in the Life of a Tree. Subscribed!

[Jay]

The implications of something like this are horrifying.

[rab2591]

These are amazing. Especially Guess I'm Dumb. I would love to hear A Day in the Life of a Tree. Subscribed!

WOW that 'Guess I'm Dumb' is great. 'Be My Baby' is off the charts great!

I can see why this would worry people - but just for the "what if" this is incredible. This stuff isn't going to replace classic Beach Boys, but for those of us who wish we could Brian take the lead on certain tracks, what a treat.

[Ebb and Flow]

I just want to say, the recent "Be My Baby" one really sounds convincingly like a classic 1964 era Brian vocal.  It even sounds double tracked the way a BB vocal from the period would have been.  It probably helps that there is a wealth of BB acapella material, both officially and unofficially released, to train a model on.

Obviously the wider implications of this tech in a wider sense are distressing for artists who are still actively recording, but in this capacity you are basically using AI tech to reproduce something that is lost (Brian's classic voice) and realizing something that never existed.  This could be used to recreate songs that Brian never recorded vocals to ("Sherry She Needs Me" should be at the top of your list) and the covers that you've posted show this could be used on songs outside of the Beach Boys canon.  It would also be interesting to hear BB songs sung by Brian with different members singing lead, Dennis singing lead on "I Get Around" etc.

[SurfRiderHawaii]

When I first saw the topic title I was excited that AL and Brian were making an album 😆

But this AI stuff is fascinating and great fun!

[HeyJude]

It's kind of fascinating, and I'm truly not singling out anybody in particular, because I honestly don't keep track of how each person has come down on any given topic, but it's pretty fascinating to see so many fans up in arms because like the remix of a track has different sounding drums, while some fans are fine with a dude and computer algorithm trying to sound like Brian Wilson.

I brought up the idea of this tech bleeding out into more joke ideas (e.g. let's hear Dennis sing "Looking Back With Love"!), but I honestly think that might be the *best* use of the tech, where there's no pretense to making it, and it's just truly for a laugh.

To the degree this tech would ever be used to "fix" or "right" something, I'd say re-singing vocals the actual BBs actually sang seems pretty low on the list. Actual missing stuff (which means we're mostly talking about "Smile" and a few odds and ends like "Sail on Sailor") strikes me as the best of a list of not-so-great applications for this. But, like, when 1976 Brian sang a lead, *that* voice is part of the whole thing. That's part of what that recording MEANS.

And, as a total aside, I find subbing in younger Brian on "Matchpoint of Our Love" particularly bizarre, as he actually sounds good on that one on the actual original recording! It's kind of infamous for that.

[FatherOfTheMan Sr101]

Fascinated to see this! Amazing work so far. Training the models can be quiiiiite time consuming!

I *may* have been working on a new Smile mix using all this fun new Ai stuff... keep your eyes and ears peeled...
There's a lot more to this AI stuff than generating new vocals, even the ability to separate specific voices and instruments can mean amazing things for good, clean new mixes of tracks we have no masters for. (Barnyard? Child is Father of the Man??? Wonderful?)

[HeyJude]

As I've mentioned, using AI to isolate pre-existing parts is very interesting/useful and sometimes revelatory, and a *very* different application of AI tech than generating new sounds/images/video, etc.

As we can also see from the recent "Revolver" remix, if this AI isolation software is used on actual multi-track sources, where some things are already isolated on high-quality sources, that isolation software is even more useful and effective.

[seltaeb1012002]

.

[FatherOfTheMan Sr101]

You're doing great work  Just keep focusing on that feeling of "oh crap, it works!"

I've been messing with models for the last few weeks whenever I have a free moment... trust me, I respect how great this has come out so far!

Ain't easy to set up!

[juggler]

I was listening to the "Feel Flows" set on a long drive today, and Carl's stunning live rendition of "This Whole World" from 1988 really knocked me out.  CW sings it with such passion and ferocious energy, that it struck me as something that an AI bot can never completely replicate. 

With that said, once the genie of any new technology is out of the bottle, then it's out of the bottle. For good or ill, if something is useful, then, by golly, it will be used.  Let's face it. If Brian had been able to use AI bots for Smile vocals instead of the Boys in 1966, many unpleasant arguments over lyrics would have been avoided.

I'm enjoying these mock-ups.  The possibilities are truly endless.

[Sam_BFC]

God Only Knows "1966 home demo" is pretty good.

Does the tag use any of the authentic vocal from the released version?

What seltaeb1012002 really should have done, is held off posting anything until April 1, and then on that day introduced the above mentioned "home demo" as a genuine long lost article. It would have been a cruel but interesting experiment and perhaps with a free pass if posting on that day 

[Sam_BFC]

Who remembers the incident during the What Love Can Do era, which also included a remake of God Only Knows, when promotional video of the new God Only Knows was posted, but apparently tampered with in some way to be part original version, part BW remake, and part assistant studio engineer's brother singing a snippet, or something like that? There was plenty of commotion trying to figure out what on earth was going on before the reveal.

[FatherOfTheMan Sr101]

AWESOME work on the latest one! So inspiring!
Can't believe this is where tech is... imagine how the geniuses of the era would feel listening to these "ghosts" 

[seltaeb1012002]

I was listening to the "Feel Flows" set on a long drive today, and Carl's stunning live rendition of "This Whole World" from 1988 really knocked me out.  CW sings it with such passion and ferocious energy, that it struck me as something that an AI bot can never completely replicate. 

With that said, once the genie of any new technology is out of the bottle, then it's out of the bottle. For good or ill, if something is useful, then, by golly, it will be used.  Let's face it. If Brian had been able to use AI bots for Smile vocals instead of the Boys in 1966, many unpleasant arguments over lyrics would have been avoided.

I'm enjoying these mock-ups.  The possibilities are truly endless.

I agree, you're only gonna get the AI to replicate whatever you feed it - with maybe some variation, depending on how much it has to go off of. At least in the iteration of the technology we have at this point.

And cool, glad you're enjoying them! I'm working on AIs for Carl, Dennis and Mike now. I'm in South Korea for 2 weeks and Google decided to lock me out of my Dae Lims account this morning without option to get back in, unless I log in from a previously known location... Enable 2-step authentication people!

God Only Knows "1966 home demo" is pretty good.

Does the tag use any of the authentic vocal from the released version?

What seltaeb1012002 really should have done, is held off posting anything until April 1, and then on that day introduced the above mentioned "home demo" as a genuine long lost article. It would have been a cruel but interesting experiment and perhaps with a free pass if posting on that day 

Aww man, that would've been perfect! Yeah, I was too caught up in the matrix to think clearly enough about how to market the thing. But whatever, the cat's out of the bag, lol.

And nope - that's all AI Brian.

AWESOME work on the latest one! So inspiring!
Can't believe this is where tech is... imagine how the geniuses of the era would feel listening to these "ghosts" 

Thank you! Yeah, I mean it's kind of a dream come true for fans. So much potential. And who knows, maybe one day these tools will be utilized for official releases. I could imagine a full AI concert experience featuring "live" vocals, holograms, etc. Imagine seeing Pet Sounds or SMiLE live, or Sgt. Pepper, the White Album, etc... Or even being able to sort of program your own concert at home. "Hey ChatGPT, play me Knebworth 1980, only replace all the songs with Love You in its entirety!"

[rab2591]

Thank you! Yeah, I mean it's kind of a dream come true for fans. So much potential. And who knows, maybe one day these tools will be utilized for official releases. I could imagine a full AI concert experience featuring "live" vocals, holograms, etc. Imagine seeing Pet Sounds or SMiLE live, or Sgt. Pepper, the White Album, etc... Or even being able to sort of program your own concert at home. "Hey ChatGPT, play me Knebworth 1980, only replace all the songs with Love You in its entirety!"

It had been my dream for years that Brian's band would do Brian Wilson Reimagines Love You - and instead of the synths (which I do love) we could hear those songs with Pet Sounds-esque instrumentals.

Perhaps, years down the road, AI could recreate the instruments/vocals on Love You to make them more of a 1965/'66 Beach Boys style. Not dying to hear this, but it would be cool if possible.
________

On another note, I find the response to these tracks to be incredible. I don't have the brainpower to decipher the varying opinions (nor the patience to read every post relating to this topic). Hey Jude seems to be right that there are a swath of people claiming they love it, and a swath of people saying they hate it, and I haven't really seen many in-betweeners.

While I have really liked what I've heard thus far, I am also hesitant about this stuff. But I also find myself hesitant to adopt a lot of mainstream modern technology, feeling nostalgia for simpler times and how people seemed to be much more inwardly/spiritually content 20+ years ago without Alexas or robot vacuum cleaners.

It's odd, because I have almost equal-opposite reactions to these AI tracks: I really am positively floored by the technology, and it's cool to be tricked into a fantasy of hearing Brian at the piano doing a demo of GoK. But I seem to also want to shut my computer down and travel back to simpler times. Something just feels wrong about it this AI stuff - like a feeling in the back of my mind - and I don't think I'm alone in feeling that.

Again, this is really cool stuff, and I'm happy that AI tech isn't just being used to help students cheat when writing essays. I love that the creatives here on the forums are delving in and trying to create something cool for us fans to get lost in. Can't shake that uneasy feeling though - though maybe that's just normal when a vastly new technology becomes available. Can't imagine what people thought of telephones after thousands of years of writing letters to one another.

[FatherOfTheMan Sr101]

First off, seltaeb, if you would like me to start my own thread I surely will. I don't wanna hijack your awesome work, but I feel it's a little redundant to create another AI thread.

Second... I have a question for you all...

Do you Like Worms?
https://youtu.be/Gdeh72-mFoA

I'm working on the full track, as well as a few more pieces of Smile. I have a separate model from seltaeb I've been training myself. I'm not perfect at performing for it yet either! That said, let me know what you think! I figure, since it isn't "real" I should add a bit of my own arrangement while keeping it true to the vision as closely as can be with the limited info and lack of recordings.

So excited to share the whole Smile mix with you all! I have a lot more going on here than just the AI vocals!

[El Molé]

I’m currently thinking of this technology as another step on from other advancements in the recording industry. Maybe that’s wrong, maybe this is fundamentally different somehow, but so many technological changes have left people feeling uneasy, or concerned that the music they’re hearing isn’t ‘real’, that I see that as a (currently) more extreme end of a spectrum.

Personally, I love the examples so far and ver much look forward to new ones. There’s clearly a lot of work involved and the results are very impressive and great listening, in my opinion. I love Brian’s voice and a lot of these examples sound very much like it.

Ethically, I’d have a problem if someone was exploiting someone using this, and the industry definitely has form in exploiting talented people, so I can see the potential for all sorts of problems there. I’d also have a problem with ‘new’ material using an AI voice, but purporting to be the original artist etc.

But the potential for using this to fix, complete or as part of a suite of tech in production could add real value, I think. E.g. as a fix for Carl’s “shunshine” flub, or to present what Dennis might have intended for some of his music when his voice had given out etc (e.g. AI Dennis plugged in to the Taylor Hawkins vocal might make the listening experience of that track amongst other Dennis tracks much smoother and more enjoyable).

What better way to test the capabilities of the tech than with the greatest unfinished album of all time? Vey much looking forward to whatever Seltaeb and others can create with AI.