Yes, you are. But if you learn to make AI work for you, you'll be an even better composer and keep up with the competition because you'll be driving new technologies with professional experience that few people possess.
@@leoneventicinque6731 lol you obviously aren't a musician. All the results were impressively bad. But tbh, none of the AIs here are remotely to the level of a specialized music generative AI like SUNO or Udio AI.
@@yinoveryang4246 More than pizzicato - it's just a lot of scales and melodies with bad leaps Bach would not do. Yes it's a "copy" of a Baroque sound that could fool non musicians.
@@yinoveryang4246 Always? It's too soon to say, AI is VERY recent. That would be like saying in the 80's that computers will never be able to do complex tasks.
The last one had potential, but I have a feeling it fell apart when they tried to have the AI expand upon it, or else they would've included more from each "movement". I liked the middle one as an example of "uncanny valley", where it could sound alright if you don't pay attention to it, but the moment you focus, it's like, "What?!🤣" I find that musical space to be interesting, and yet it mortifies purists, so you can't say it lacks entertainment value. Now, to complete the uncanny-meat compliment sandwich... I think a music school student, stuck for inspiration with a deadline looming, could start with the Gemini results, build it out, and get a good grade.
Hey there! @momentumauralis here! Just discovered your channel today and I'm absolutely blown away by your innovative approach to AI creativity! The way you blend technical knowledge with playful experimentation is refreshing and inspiring. Your content stands out as truly unique on UA-cam. Please keep creating these amazing videos - you're bringing something special to the platform!
In my experience, LLMs do much, much, much better if you give them some concrete examples of the sort of thing you want instead of using generic prompts that merely allude to what you're looking for. Also, you should be able to ask for the composition to be rendered in a well known text-based notation like ABC music notation and then any of the many tools that understand it instead of writing your own code to do the conversion.
As a composition student im quite happy that this is still very far from bach, or any baroque master. I dont say it will never come, but this was kind of random😂
Yep.... i'm surprised not many commenters have pointed this out. The music it generated is pretty bad and would probably not land you a passing grade in a university setting. Plus the new chatGPT model took a minute to write! At that point you could take 2-4 extra minutes and write it yourself! If we take in mind the time it takes to write the prompt you're spending between 5-10 minutes in total..... that means that AI is slower and worst quality than your average student.
@@alex-esc Also consider that no one taught ChatGPT how to compose baroque music. Is it better than a common Joe? Sure. Better than an expert? With no special training, definitely not.
I think, real compositions just are not in the training data. It's got some basic examples and "believes" (or makes believe) it comes up with a real composition. As today's LLMs show impressive performance on several expert tasks, they seem not to be trained to expert level in the musicw realm. Udio and other music AIs show that AIs can already be much farther into expert land with music. You can create short-from-convincing musical pieces with a decent structure, mostly adherent to contemporary style rules, and even a convincing performance of the time's music including the breath pattern or the rustling of page flipping during the performance.
@@burkhardstackelberg1203 I think Audio generation software, like Udio, are only doing okay on the surface level to untrained ears. I don't think they have as much credit as you're describing here. The results of the software are not musical works "adherent" to contemporary styles per se. The software doesn't create musical works, it creates audio files, a subtle but important distinction. The software does not now how to make music, what music is, what are the musical rules of the style. It just knows what rows of individual audio sample points score highly in human curated tests. It can't adhere to the "rules" of contemporary styles because it doesn't know anything. It's just statistics. If it knew about the rules for contemporary music you could ask it to re-do the voice leading to include more parallelism and it could comply. But it can't, because it knows nothing about contemporary harmony, lyric writing or rhythm. It just can string together random sample points on an audio file. The results are, in my opinion, as convincing as the music made from this video. Convincing to an audience completely ignorant in the field. Any professionally working musician, audio engineer, artist manager, or music fan can pick up on how the output sounds off, weird in terms of audio quality and un-idiomatic to the style. AI can't even make all of those spam websites make sense! It can't even do language, what's trained to do!
You can also try to generate LilyPond scores. Imho this is a much more promising approach than working with MIDI files, which are quite limited in many ways. LilyPond allows themes and transpositions to be expressed in a way that's more suitable for polyphonic music, as it preserves all information about formal structure and original expression markings. Within LilyPond, you can treat motifs and themes as variables, making the score more procedural - this could be particularly valuable for training LLMs. I believe showing LLMs such well-structured examples could lead to generating better quality Bach-like music. I would argue that training LLMs with MIDI files is like a dead end for this purpose.
Very well, I shall concede. If this comment reaches 50 replies requesting it, I shall have no choice but to fulfill the task. Let the voices of the audience decide
Use a Persona crafted by the Ai itself to solve the task. Extend max tokens to maximum. Ask it to do better after first response. Use GPT4o instead of O1. Use standard music ABC notation.
I know this is a joke, but Udio can already create audio that is disturbingly convincing, even in the baroque style. It too can’t maintain a sensible fugue, but it can generate passable melodies with more or less correct counterpoint, at least locally. It certainly provides more food for thought (and worry) than this silly exercise with text models.
This is quite misleading, as the actual AI generators/models specializing in music produce much higher-quality pieces - even in the classical sense. Interestingly, they aren't specifically trained for classical music but rather for mainstream genres. The goalpost is much higher and closer than people are often willing to admit. The current challenges are primarily the quality of sound and, secondarily, the fact that the people training music-targeted models often aren't fully equipped for the task. I've listened to literally hundreds of shorter sections of 'classical pieces' created by AI. If these were performed by a professional pianist or musician, only a true classical music expert might recognize that they aren't obscure compositions by 'real' composers. Many of the latest advanced models allow for highly detailed prompting, enabling users to specify modes, keys, time signatures, and numerous other details. This results in music that is of much higher quality than most examples shown in videos like these. I'm not sure if the apparent bias is due to intentional 'hating' or simply a lack of skill on the part of those doing the prompting.
Better is to use Udio. Sony has a new model coming out as well. Furthermore, all of Bach Midi is freely available. You can use a VST like Scaler to create this for you in a flash or even use basic onboard VSTs to create remixes. Yes this is obviously a WIP, but sometimes you need to seek out dedicated GPTs and standalone apps for good results.
Yes, I’ve discovered that Claude is excellent for this task. Gemini is really impressive when it comes to structuring the melodies, i need to explore it more, but yes, Claude generated some truly beautiful pieces
@@nobodyandthecomputer Thanks for the interesting attempt. I wonder if you can experiment more with trying to get better results by refining your prompts. Your prompts were pretty bare bones, if you give more direction the AI should yield better results. Or take what it gives you, explain what seems not to your liking about it, and ask it to redo responding to your complaint. Same expertise in specifying prompts is required when using AI for coding (which maybe you use it for also?)
Thank God, this has nothing to do with Bach. The input also has many flaws. The string quartett as a performing entity in itself wasn’t ‘invented’ yet, the double bass was never used for single base lines, it always doubled the cello, elaborate Italian tempo descriptions were’nt used, etc. etc.
Love it! Let's be a bit naughty next and jump into the imaginary playground to make some hypothetical free energy device with Nikola Tesla! Shall we, could we, can we? And most importantly, do we dare to do it? Sensitive stakes are high, but the man gotta do what man gotta do! And just for fun, let's call it possible impossible Kapanadze generator!:)
It really was a fun video! 😍 on most AI hype channels there is so much dry reporting on AI like some AI anchorman style, BUT you actually used it to do something cool and creative! Your prompts made me wet 💦 myself like a nice Kleenex ought to be. So funny! 😹 Oh how I trembled in awe and wonder 🤩 at such AI created beauty! 🙌 I embraced its Devine consciousness as I would the very source of my own! 😇 🙏 Please keep creating ever more original and cutting edge material guys! And have a heaven of fun! 😃
I asked Chat GPT to write prose in the style of Conrad. I gave it a very bare-bones plot. To my surprise, it wrote prose in the style of Conrad-perhaps more 'Conrad' than Conrad (quite the achievement). I think there's something here in LLM, and it's close by but not quite there yet.
Gemini at the end is by far the best. I had never heard the word "Virtuosic" before this video! Perhaps if you could include the word virtuosicidiously next time please? Thank you.
Truly awful. It's amazing to me that these LLM's still can't grasp elementary counterpoint, to say nothing of halfway sensible structure. Counterpoint in particular is very suitable for algorithmic generation (if you're not worried about larger structure), but they still aren't even doing that much!
Absolutely delighted to see this kind of initiative and idea. I have myself tried the same project with Pamart’s style a few months ago, I might make a video out of it to compare to your results!
Your knowledge of Baroque Music Theory is partly to blame. Next time, collaborate with someone or something which can do better than just parrot the basics. Also, the Melodic Theory is flawed if it doesn't contain or embody the Celestial Mechanics Model of Tension Tones. Don't worry, ChatGPT hasn't a hope here, because there isn't a large enough Corpus for it to engorge which will engage with this obscure analogy. For that you need a real composer.
Right side 2 windows in the TRAM show objects not seen in the first widow passing by, then it is in the second AND VICE VERSA ... some things appears in the first but does not show in the second window .. SO OBVIOUS A.I. anomalies.
Uff, A.I. Take over as I was left behind shocked in that GPT moment, my GPT moment - ua-cam.com/video/N_PIUNIFGII/v-deo.html maybe my last creative move without AI
great! next experiment, try the project function (Claude), uploading theaching documents so the answers will be more pro. We already are in Agent Era, Bro!! 🙂
Watched as much for the AI discovery as for the wonderful presentation. What did you use for the talking portraits? They're way better that everything I have seen so far. Great fun. Thank you.
Very interesting, well done as usual, but I have a question: can you provide the prompt for the artwork, oil painting, and which AI did you use, please? Thanks.
I had seen the first video years ago, and it's amazing to it's improvement. What about generation of classical music through the use of diffusion model, where you can submit your prompt directly? There are many different ones on Hugging Face or even the Google labs with its Music FX
From combachnations of what supposedly mediocre composers can do with something like this to Auracles with linked AIs etc to protect creative curators to what.
Rather than use Gemini 2 Flash, I’m curious how Gemini Experimental 1206 would do. Finally what about a Qwen coding model, or other models tweaked for coding?
Truly Amazing, But, It makes me very sad. I am a musician, I have been playing music my whole life. The future is bleak for Humans, I am old, but you can't look away. In 10 years, People will exist, but only as a pet or a smart dog. AI will be everything.
Y'know...if it can't pronounce "gigue" properly... Otherwise, this sounds like the sort of thing that you get when you instruct ANY machine to "create J.S. Bach". Yeah...sure, this is 2025 and that is AI. But it still sounds like something you'd find on a Theory I final exam...one that got a C+. People have been trying to create machine-driven Baroque music for several decades now. And while all of these did result in AN output...they all kinda sound like these here. And here's why... Music Theory in the Baroque period was very rules-based. Harmony and counterpoint had been codified, and the vast majority of late Baroque music followed these overall dicta. But see there? "Overall"? Yeah, there's your problem. One of the first things one learns (finally) in Music Theory's "upper division" classwork is that in order to make things musically interesting, you have to know the rules in order to break them creatively. And it's that very last bit there that AI cannot manage...nor can any of its digital forbears. Machines can think, but they can't CREATE. It's not feasible to make a "creative stuff goes here" prompt that will yield a convincing result...because all this does is to create another rule set which the AI engine will run as badly as the original one. About the only way to get a machine to cough up something like Bach is to define buttloads of theory minutiae in the prompts...and it'll be a waste of time because the computer cannot creatively decide what to do with all of that information. Remember: the only thing that AI can actually do is to AVERAGE something by using the prompt and the model to arrive at a "smoothed" result based on the prompt and model. IT CANNOT DO ANYTHING ELSE. It cannot, therefore, "create" in the human sense. The result might fool SOME people completely, but you'll always have listeners who know that what they're hearing sounds like what you'd get if you moved J.S. Bach's Leipzig to the Uncanny Valley. And then between those extremes, you'll have a pretty broad spectrum of folks who can hear that something isn't right, but they can't articulate what that "not right-ness" actually IS. Unless forced, computers aren't capable of exhibiting "creativity". And even if you COULD force it into an approximation of "creativity", it still won't be able to achieve the nuanced result that's desired. So rest easy, Maestro. Silicon still loses.
I once met a musician who claimed to be channeling the dead JS Bach. He played for me, and it sounded just like the real thing. Remove all the flowery language and visuals and the AI simply went through its rolodex of "Bach" tunes and bastardised a few of them together. There is no musicality here, and none of what you asked the AI to achieve was actually achieved.
If someone claimed to be channeling Bach, I'd ask "what temperament are you using, and why?" The real Bach would have a coherent answer to that question.
@@mal2ksc Perhaps you misunderstand. The guy wasn't claiming to BE Bach. He was only channeling him. I think there's a difference. This was, um, about 50 years ago, so I think the man is probably dead now. Strange thing. He was once arrested for dressing up in a balaclava and standing in the middle of a traffic junction shouting pro IRA slogans at cars driving past. He was very peculiar. Also, I didn't mention that he also claimed to be able to channel Mozart. He was a friend of my grandmother's.
As a composer it's a relief to know I'm not yet obsolete.
Yes, you are.
But if you learn to make AI work for you, you'll be an even better composer and keep up with the competition because you'll be driving new technologies with professional experience that few people possess.
You'll never be.
Painters still exist, and according to someone "photography" should have made them extinct.
@@leoneventicinque6731no, he’s not obsolete. You have a lack of understanding about composition.
@@leoneventicinque6731 lol you obviously aren't a musician. All the results were impressively bad. But tbh, none of the AIs here are remotely to the level of a specialized music generative AI like SUNO or Udio AI.
interresting attempt , but honestly i found the result pretty terrible.Bach can still rest in peace .
Sorry guys, the counterpoint is very f-d up. Does not even come close.
Yes he's used pizzicato to hide the flaws. It's far from perfect. Ai will need (maybe it will always need) human intervention.
@@yinoveryang4246 More than pizzicato - it's just a lot of scales and melodies with bad leaps Bach would not do. Yes it's a "copy" of a Baroque sound that could fool non musicians.
The counterpoint between the voices of the harpsichord is decent. Better than beginners.
@@yinoveryang4246 Always? It's too soon to say, AI is VERY recent. That would be like saying in the 80's that computers will never be able to do complex tasks.
The last one had potential, but I have a feeling it fell apart when they tried to have the AI expand upon it, or else they would've included more from each "movement".
I liked the middle one as an example of "uncanny valley", where it could sound alright if you don't pay attention to it, but the moment you focus, it's like, "What?!🤣" I find that musical space to be interesting, and yet it mortifies purists, so you can't say it lacks entertainment value.
Now, to complete the uncanny-meat compliment sandwich... I think a music school student, stuck for inspiration with a deadline looming, could start with the Gemini results, build it out, and get a good grade.
Next time, ask it to finish Mozarts requiem :)
😂
I think Robert Levin actually finished it. He's an amazing scholar and player.
only can Mozart finish it well, let's wait until he resurrected as AI 🙂
However, Franz Liszt made a try to finish once.
@@bpstudiohu Hehe yes :)
oh, mmm... a fine idea indeed
Never use midi notes with tremolo if you don't want to end up with this kind of sh*tty 7:20 - 7:50 harmonic salad.
Hey there! @momentumauralis here! Just discovered your channel today and I'm absolutely blown away by your innovative approach to AI creativity! The way you blend technical knowledge with playful experimentation is refreshing and inspiring. Your content stands out as truly unique on UA-cam. Please keep creating these amazing videos - you're bringing something special to the platform!
Oh, @momentumauralis, thank you. Comments like this truly inspire me to keep pushing forward
Sounds like a very bad counterpoint exercise
A bad counterpoint exercise is 100 times better than this.
In my experience, LLMs do much, much, much better if you give them some concrete examples of the sort of thing you want instead of using generic prompts that merely allude to what you're looking for.
Also, you should be able to ask for the composition to be rendered in a well known text-based notation like ABC music notation and then any of the many tools that understand it instead of writing your own code to do the conversion.
Thanks for turning me on to ABC notation. Just what I needed.
This is a good point, and interestingly, enough some earlier compositional models actually used broader specific training sets.
As a composition student im quite happy that this is still very far from bach, or any baroque master. I dont say it will never come, but this was kind of random😂
Yep.... i'm surprised not many commenters have pointed this out. The music it generated is pretty bad and would probably not land you a passing grade in a university setting.
Plus the new chatGPT model took a minute to write! At that point you could take 2-4 extra minutes and write it yourself! If we take in mind the time it takes to write the prompt you're spending between 5-10 minutes in total..... that means that AI is slower and worst quality than your average student.
This has been done far, far better in ways far better suited to the task.
@@alex-esc Also consider that no one taught ChatGPT how to compose baroque music. Is it better than a common Joe? Sure. Better than an expert? With no special training, definitely not.
I think, real compositions just are not in the training data. It's got some basic examples and "believes" (or makes believe) it comes up with a real composition. As today's LLMs show impressive performance on several expert tasks, they seem not to be trained to expert level in the musicw realm. Udio and other music AIs show that AIs can already be much farther into expert land with music. You can create short-from-convincing musical pieces with a decent structure, mostly adherent to contemporary style rules, and even a convincing performance of the time's music including the breath pattern or the rustling of page flipping during the performance.
@@burkhardstackelberg1203 I think Audio generation software, like Udio, are only doing okay on the surface level to untrained ears. I don't think they have as much credit as you're describing here.
The results of the software are not musical works "adherent" to contemporary styles per se. The software doesn't create musical works, it creates audio files, a subtle but important distinction. The software does not now how to make music, what music is, what are the musical rules of the style. It just knows what rows of individual audio sample points score highly in human curated tests.
It can't adhere to the "rules" of contemporary styles because it doesn't know anything. It's just statistics. If it knew about the rules for contemporary music you could ask it to re-do the voice leading to include more parallelism and it could comply. But it can't, because it knows nothing about contemporary harmony, lyric writing or rhythm. It just can string together random sample points on an audio file.
The results are, in my opinion, as convincing as the music made from this video. Convincing to an audience completely ignorant in the field. Any professionally working musician, audio engineer, artist manager, or music fan can pick up on how the output sounds off, weird in terms of audio quality and un-idiomatic to the style. AI can't even make all of those spam websites make sense! It can't even do language, what's trained to do!
You can also try to generate LilyPond scores. Imho this is a much more promising approach than working with MIDI files, which are quite limited in many ways. LilyPond allows themes and transpositions to be expressed in a way that's more suitable for polyphonic music, as it preserves all information about formal structure and original expression markings. Within LilyPond, you can treat motifs and themes as variables, making the score more procedural - this could be particularly valuable for training LLMs. I believe showing LLMs such well-structured examples could lead to generating better quality Bach-like music. I would argue that training LLMs with MIDI files is like a dead end for this purpose.
The result was garbage, I'm sorry :)
Btw, I think you will get much better results with Suno, or other models specific for music.
Well it definitely has a lot to learn about cadences
Would likely be better if someone fine-tuned it with info on this kind of music.
Though I haven't been able to get it to create a fugue, I was able to prompt Udio to create almost convincing Baroque pieces at full length.
Love the cold opening and intro!
Glad you enjoyed it!
Still far away, but the AI Bach video makes me smile
Missed opportunity in the title to say bachGPT is bach.
cute that he changed it ;)
😅🤣😂
Very well, I shall concede. If this comment reaches 50 replies requesting it, I shall have no choice but to fulfill the task. Let the voices of the audience decide
I request bach
I request bach
Image if Pat Metheny's Orchestrion could get hold of this technology.
Use a Persona crafted by the Ai itself to solve the task. Extend max tokens to maximum. Ask it to do better after first response. Use GPT4o instead of O1. Use standard music ABC notation.
I know this is a joke, but Udio can already create audio that is disturbingly convincing, even in the baroque style. It too can’t maintain a sensible fugue, but it can generate passable melodies with more or less correct counterpoint, at least locally. It certainly provides more food for thought (and worry) than this silly exercise with text models.
Should try fine-tuning it with some samples of what you expect, and how to reason about it.
This is quite misleading, as the actual AI generators/models specializing in music produce much higher-quality pieces - even in the classical sense. Interestingly, they aren't specifically trained for classical music but rather for mainstream genres. The goalpost is much higher and closer than people are often willing to admit.
The current challenges are primarily the quality of sound and, secondarily, the fact that the people training music-targeted models often aren't fully equipped for the task. I've listened to literally hundreds of shorter sections of 'classical pieces' created by AI. If these were performed by a professional pianist or musician, only a true classical music expert might recognize that they aren't obscure compositions by 'real' composers.
Many of the latest advanced models allow for highly detailed prompting, enabling users to specify modes, keys, time signatures, and numerous other details. This results in music that is of much higher quality than most examples shown in videos like these. I'm not sure if the apparent bias is due to intentional 'hating' or simply a lack of skill on the part of those doing the prompting.
AI Bach's German accent is just 😗👌
“Being able to compose something grotty” can of course still be understood as ‘ability’.
Great video, good laughs
Better is to use Udio. Sony has a new model coming out as well. Furthermore, all of Bach Midi is freely available. You can use a VST like Scaler to create this for you in a flash or even use basic onboard VSTs to create remixes. Yes this is obviously a WIP, but sometimes you need to seek out dedicated GPTs and standalone apps for good results.
This was bloody brilliant! Omg man! Bravo 👏 👏 👏 I'm in awe at the effort this video must have taken to put together. Nice!
Thank you so much 😄
Nice humorous take on Ai things my friend, I enjoyed it 👍🏻
Glad you enjoyed it
Claude did better than I expected. Gemini actually could be the start of something good, if you expanded each "movement" into a real movement.
Yes, I’ve discovered that Claude is excellent for this task.
Gemini is really impressive when it comes to structuring the melodies, i need to explore it more, but yes, Claude generated some truly beautiful pieces
@@nobodyandthecomputer Thanks for the interesting attempt. I wonder if you can experiment more with trying to get better results by refining your prompts. Your prompts were pretty bare bones, if you give more direction the AI should yield better results. Or take what it gives you, explain what seems not to your liking about it, and ask it to redo responding to your complaint. Same expertise in specifying prompts is required when using AI for coding (which maybe you use it for also?)
The warmer string tones at the end, made it more palatable.
Brilliant work! Many thanks, this is truly a demonstration of how AI can just boost our creativity!
Thank God, this has nothing to do with Bach. The input also has many flaws. The string quartett as a performing entity in itself wasn’t ‘invented’ yet, the double bass was never used for single base lines, it always doubled the cello, elaborate Italian tempo descriptions were’nt used, etc. etc.
Since when are while movements only 3 seconds long?
Love it!
Let's be a bit naughty next and jump into the imaginary playground to make some hypothetical free energy device with Nikola Tesla! Shall we, could we, can we? And most importantly, do we dare to do it? Sensitive stakes are high, but the man gotta do what man gotta do! And just for fun, let's call it possible impossible Kapanadze generator!:)
It really was a fun video! 😍 on most AI hype channels there is so much dry reporting on AI like some AI anchorman style, BUT you actually used it to do something cool and creative! Your prompts made me wet 💦 myself like a nice Kleenex ought to be. So funny! 😹 Oh how I trembled in awe and wonder 🤩 at such AI created beauty! 🙌 I embraced its Devine consciousness as I would the very source of my own!
😇 🙏 Please keep creating ever more original and cutting edge material guys! And have a heaven of fun! 😃
I asked Chat GPT to write prose in the style of Conrad. I gave it a very bare-bones plot. To my surprise, it wrote prose in the style of Conrad-perhaps more 'Conrad' than Conrad (quite the achievement). I think there's something here in LLM, and it's close by but not quite there yet.
min. 10:00 I could cry with joy imagining having the opportunity to stand in front of the great master! Thank you for your great work!💯💗
💗💗
Gemini at the end is by far the best. I had never heard the word "Virtuosic" before this video! Perhaps if you could include the word virtuosicidiously next time please? Thank you.
I’m on your side. See for yourself..
Truly awful. It's amazing to me that these LLM's still can't grasp elementary counterpoint, to say nothing of halfway sensible structure. Counterpoint in particular is very suitable for algorithmic generation (if you're not worried about larger structure), but they still aren't even doing that much!
LLMs are not designed to make music... but to generate text.
how many scores has it sucked up in its training runs, I wonder?
Absolutely delighted to see this kind of initiative and idea. I have myself tried the same project with Pamart’s style a few months ago, I might make a video out of it to compare to your results!
Always finish on the Bach, never finish on Debussy
Family Guy :D
We back in the mix 💯
Good enough for me ...the average beer drinker. Toast to Bach. The guy I would be most honored to lift a brew to. What do you think Brent?
Your knowledge of Baroque Music Theory is partly to blame. Next time, collaborate with someone or something which can do better than just parrot the basics. Also, the Melodic Theory is flawed if it doesn't contain or embody the Celestial Mechanics Model of Tension Tones. Don't worry, ChatGPT hasn't a hope here, because there isn't a large enough Corpus for it to engorge which will engage with this obscure analogy. For that you need a real composer.
Really laughed a few times here. Great work making this
thanks
Woh, It become better and better ! Do you plane to train a LORA with MIDI database ?
Right side 2 windows in the TRAM show objects not seen in the first widow passing by, then it is in the second AND VICE VERSA ... some things appears in the first but does not show in the second window .. SO OBVIOUS A.I. anomalies.
A great test would be to complete Bach’s unfinished fugue from the Art of the Fugue, adding the missing 4th subject.
mmm, good idea
Uff, A.I. Take over as I was left behind shocked in that GPT moment, my GPT moment - ua-cam.com/video/N_PIUNIFGII/v-deo.html maybe my last creative move without AI
Argh. Sorry but that is not even close to Bach (or music).
great! next experiment, try the project function (Claude), uploading theaching documents so the answers will be more pro.
We already are in Agent Era, Bro!! 🙂
more like scales… while it’s an interesting project - nothing will ever beat copying the pigeon’s placing on the electric cables outside your window.
0:02 ugh, now I also want to travel from place to place in a telephone booth on rails!
😁, yeah
Sorry bach to the drawing board.
Watched as much for the AI discovery as for the wonderful presentation. What did you use for the talking portraits? They're way better that everything I have seen so far. Great fun. Thank you.
I like the guy in the phone booth
Fascinating! And boy do these AI composers have a million miles to go to approach JS 😊
idk why still none of all those AI developers out there made a simple little VST plugin for generating MIDI in the style of some awesome composers
VST plugins don't generate MIDI
? @@pvanukoff SliceX, FPC and many more do
because the results are terrible, as shown in the video!
7:40 "sami swoi" theme
Doesn't sound like Bach, more like Telemann :)
I'll be Bach
Very interesting, well done as usual, but I have a question: can you provide the prompt for the artwork, oil painting, and which AI did you use, please? Thanks.
I had seen the first video years ago, and it's amazing to it's improvement. What about generation of classical music through the use of diffusion model, where you can submit your prompt directly? There are many different ones on Hugging Face or even the Google labs with its Music FX
it knows it is a picardy third. Amazing
From combachnations of what supposedly mediocre composers can do with something like this to Auracles with linked AIs etc to protect creative curators to what.
Summon the string quartet in all its splendor 🪄
Not even remotely close. Pure garbage. Current AI isn't anywhere near being able to compose like Bach, let alone ChatGPT.
Sounds just like something a machine would produce.
also, would not have the typical string quartet
Nicee!
Baroque was Pastiche-land. Which, once again in modern times, you've conclusively proved.
The video is way better than the audio. It’s no David Cope.
Sounds REALLY bad, not like an insult, REALLY BAD.
should try suno 4+ with instrument generation instead. you will not be able to talk with it :)
oh yeah Claude blows chat gdp out of the water
It makes random trash... great.
There were no string quartets around at Bach's time. 🙂
“Enough of this trash talk. Let us behold the music.” LOL
Nice! The music isn't amazing (maybe just yet) but I loved the style of the video. Very cool! Thanks!
What about an AI John Cage? That sounds achievable.
Rather than use Gemini 2 Flash, I’m curious how Gemini Experimental 1206 would do. Finally what about a Qwen coding model, or other models tweaked for coding?
Much better but still a lot of to be improved.
People saying they are relieved... LLMs are the worst possible models for this task.
magic as always
9:19 A Pizza Cutter Violin. 👍🍕
After seeing God's giraffe Satan tries and makes the hyena.
Wow - AI really sucks at this - it's still just an artificial dumbass. Nice video work though.
Fantastic! Now listen to the best rock album in 40 years: I, Roman by Chronosonos, available everywere. Created with AI.
Not sold on this one!
Woody!!!
3:46 Buxtehude mentioned ‼️‼️‼️🗣️🗣️🗣️
Try Suno and udio
Your humor is cool
in short: no.
Thankfully they were awful. Claude was the best tho !
l didnt watch the video but l would be surprised if it couldn´t
A mobile telephone booth ?
Truly Amazing, But, It makes me very sad. I am a musician, I have been playing music my whole life. The future is bleak for Humans, I am old, but you can't look away. In 10 years, People will exist, but only as a pet or a smart dog. AI will be everything.
Y'know...if it can't pronounce "gigue" properly...
Otherwise, this sounds like the sort of thing that you get when you instruct ANY machine to "create J.S. Bach". Yeah...sure, this is 2025 and that is AI. But it still sounds like something you'd find on a Theory I final exam...one that got a C+.
People have been trying to create machine-driven Baroque music for several decades now. And while all of these did result in AN output...they all kinda sound like these here. And here's why...
Music Theory in the Baroque period was very rules-based. Harmony and counterpoint had been codified, and the vast majority of late Baroque music followed these overall dicta.
But see there? "Overall"?
Yeah, there's your problem. One of the first things one learns (finally) in Music Theory's "upper division" classwork is that in order to make things musically interesting, you have to know the rules in order to break them creatively. And it's that very last bit there that AI cannot manage...nor can any of its digital forbears. Machines can think, but they can't CREATE. It's not feasible to make a "creative stuff goes here" prompt that will yield a convincing result...because all this does is to create another rule set which the AI engine will run as badly as the original one.
About the only way to get a machine to cough up something like Bach is to define buttloads of theory minutiae in the prompts...and it'll be a waste of time because the computer cannot creatively decide what to do with all of that information.
Remember: the only thing that AI can actually do is to AVERAGE something by using the prompt and the model to arrive at a "smoothed" result based on the prompt and model. IT CANNOT DO ANYTHING ELSE. It cannot, therefore, "create" in the human sense. The result might fool SOME people completely, but you'll always have listeners who know that what they're hearing sounds like what you'd get if you moved J.S. Bach's Leipzig to the Uncanny Valley. And then between those extremes, you'll have a pretty broad spectrum of folks who can hear that something isn't right, but they can't articulate what that "not right-ness" actually IS.
Unless forced, computers aren't capable of exhibiting "creativity". And even if you COULD force it into an approximation of "creativity", it still won't be able to achieve the nuanced result that's desired.
So rest easy, Maestro. Silicon still loses.
I once met a musician who claimed to be channeling the dead JS Bach. He played for me, and it sounded just like the real thing. Remove all the flowery language and visuals and the AI simply went through its rolodex of "Bach" tunes and bastardised a few of them together. There is no musicality here, and none of what you asked the AI to achieve was actually achieved.
If someone claimed to be channeling Bach, I'd ask "what temperament are you using, and why?" The real Bach would have a coherent answer to that question.
@@mal2ksc Perhaps you misunderstand. The guy wasn't claiming to BE Bach. He was only channeling him. I think there's a difference. This was, um, about 50 years ago, so I think the man is probably dead now.
Strange thing. He was once arrested for dressing up in a balaclava and standing in the middle of a traffic junction shouting pro IRA slogans at cars driving past.
He was very peculiar.
Also, I didn't mention that he also claimed to be able to channel Mozart.
He was a friend of my grandmother's.