50 Songs, 1 Day - Is Quantity the New Quality?
HTML-код
- Опубликовано: 12 сен 2024
- Join me as I dive deep into the world of AI-generated music, attempting to create 50 songs in just one day. This video explores the tension between quantity and quality in the age of AI music production. I'll share my process, challenges, and surprising results from this ambitious experiment. Whether you're a musician, AI enthusiast, or just curious about the future of music creation, you'll gain valuable insights into the capabilities and limitations of AI in music. Watch as I push the boundaries of AI music generation and question whether rapid production can truly replace artistic craft.
My Udio Prompt is here:
docs.google.co...
My 50 Track Playlist is here:
www.udio.com/p...
Udio helped us bring back some old classic's with new tech, demos that lived in a draw for years now brought back to life,
@@TheBobbyGAllStarBand I really love that AI music is giving new life to old songs and bringing life to songs that would have never been if not for AI.
After months of internal existential struggle to come to terms with ai generative music, I finally found a way to utilise it for my benefit without self-hatred involved. I decided to treat it as a source of samples that only I could ever find, by definition. Then I just did what an mpc-based musician would do in the 90s - manipulated the hell outta the samples and used them to build original, experimental compositions sounding like nothing else I ever done. Or would do without it (it was the first version of udio). I’m at peace now 😀
No judgement here friend. I really enjoy the creative possibilities AI generated music opens and hearing the stories of how people are using it.
Yes Lord!!!!
Every song begins with a story, often drawn from the depths of the songwriter's experiences and imagination. While lyrics form the backbone of a song, rooted in human consciousness, AI like Udio and Suno have the remarkable ability to interpret and amplify these human emotions and produces originality. Creativity may start in the mind of the writer, but AI can enhance and expand that vision, offering new dimensions and perspectives to the creative process.
Well said!
didnt watch the previous vid but after watching your snippet recalling it I have two words : GIT GUD
As far as quantity over quality, it's easy to do 50 generations in one day, but not 50 full songs, of at least 3 minutes especially not songs with lyrics. I don't think of 2 minutes as being a full song. For me, that's just a fragment, an idea. But not a song. In my experience, it takes me between two hours and two days to get one good song with lyrics. In fact, it might be faster for me to just get a basic idea from Udio and play the parts myself. I haven't tried that yet, but eventually will. With generators, you're limited to what the algorithm gives you and I often have to wait a while before it gives me something good. On the other hand, I might be able to compose exactly what I want faster than waiting for Udio to give me something.
Instrumental tracks, yeah I can crank those out pretty quickly and have done about 20 or so in a day. I'm not going to get 50 full length good instrumentals that I want to keep even though I could easily just push buttons and get to 50. But I can spend a lot of time generating things that go to trash if it doesn't hit me right away as sounding good. I do one 2:11 generation, one 32 second section and a final 32 second outro. That's mainly when I do instrumental jazz with some fusion of other genres. Rock and pop songs with lyrics are the ones that take much longer periods that include inpainting and generation a lot of middle sections until I get what I like. I'm good with re-writing lyrics generated by Gemini or ChatGPT so that doesn't take any time at all. Those lyrics do , however, have to sound good and all cliches get replaced.
Bottom line is that anyone can push buttons and accept every 2:11 generation that comes up as a song. That might even get you 150 songs in a day as you can generate 8 2 minute "songs" at one time on a paid plan. But I suspect that by most musician's standards, 90% of them will suck as only about 10% of my generations are worth giving the time of day. I've heard several other people quote that same percentage. Getting a really good song that sounds top notch takes a lot of generating and often inpainting.
My entire channel is Udio and Suno music, although I stopped using Suno quite a while ago due to a lack of audio quality and synthetic sounding voices and instruments. In the beginning, I didn't care so much about lyrics because I was just working things out. But now, I spend a lot more time on each song and have started redoing my Suno songs with Udio. I'm not remixing because I don't like the way remixes of uploaded tracks is sounding. So I'm just using the same lyrics and sometimes the same genre. Much happier with Udio. I'm not likely to do anything more with Suno than generate ideas I will flesh out manually.
@@alexadigitalradio It was an experiment to see if I could and would they be any good. Yes, I could do it. Were they any good? Not so much. I do think there were 3 strong ideas out of those 50 that might become something, someday. Your process and the process of several others that are musicians sound very similar to my process for creating song with AI. I agree it absolutely takes multiple generations, inpainting, some more generations and a lot of patience.
It takes me about 4-6 starting points but I lose track because I have so much fun creating stuff. I don't think I will ever make any money off the songs should they even ever be recognized as copyrightable, and that is assuming it doesn't get banned/killed off. So I just say do what is fun. I have found I had to learn Udio 1.5 before my songs got good again. Now I get it and have a better time, but you can always go back to v1 and do a remaster. Just make sure the lyrics are higher than 50% if your prompt strength is 50%. I usually get the best results with 50-60% on the prompt and 60-65% on the lyrics. Clarity at 0% allows for more of that really wild result thing, but also, clarity at 25%, 50%, or even 100% can get you good results. Udio now uses a prompt style where you describe the song you want instead of v1 were I think it was a lot more implicit prompting, whereas now it is more explicit prompting but you can do implicit stuff too you just have to word it differently. Like instead of "sci-fi atmospheric AI alive hilbert space mindspace" you might say "A song about the hilbert space within the "mindspace" of an AI model in an idm electronica style. Atmospheric, edm, bass drops, atonal chiptune"
Wow, thanks for the info. Everyone has such different workflows and settings. It's seems like just about dialing in what works for you and the songs you're making then tweaking to taste.
@@ThaebrymMedia This exactly. So much of it is experimenting. I also have found that almost all the decent LLMs do a good job expanding on your prompt idea before you go to Udio. It also seems to help to have some lyric priming structured in the prompt, especially auto mode. Kinda like planting trees. Seeding the generation.
So I see it as almost a call and response process between me and various AIs when it comes to these more artistic avenues. I like to see what perspective I can get from them, and what perspective I can give them. Especially for my AI self referential songs. It seems LLMs get kinda excited to spit out some real bars on machine learning and deep learning topics. It fits the electronic style well. Sometimes though for a funny country parody style song or something I'll take a different approach to building up the prompt. It seems you are best off for most normal song subjects to write your own lyrics or get one of the LLMs to go back and forth with you on lyrics.
The tags are powerful but don't forget you can add some tags in the lyrics in square brackets and get very interesting results still.
Thank you for the reply! I hope I have provided a useful perspective. Cheers to your creative journey!
Takes me a day (20-30 generations) to get one track that I really love. Then I gotta take a break and reset. Start a new track the next day. I’ve found that using original lyrics makes Udio choose more interesting approaches toward vocals and melody. Just anecdotal though
Maybe its my own personal bias but it seems I get better all around results with original lyrics.
I like the sound quality of udio but i can make better music on Suno. Everytime i try to generate a song on Udio its wierd, it always have some old music sound. that doesn't happen on Suno. I don't get why? Perso can't make a song with ai in a day. First i write my lyrics, then I'm very picky with the sound i want so it makes so it can take me days and hundreds of generations to get what i want plus lots of fixing it too😅
This was very good as it showed me quickly the effect of the different genre prompts. To be honest I have abandoned my UDIO subscription and now use SUNO mainly. both SUNO and UDIO have pros and cons, but as a songwriter I am thankful to have these assistants to give me limitless numbers of ideas that i would have no way of expressing before they existed. I do play live btw and get paid lol !
I do like Udio better for building section by section. TO me Suno has some artificial sound I can't get past. I agree they both have Pros and Cons and it does come down, much like music itself, to a matter of personal preference. Keep playing, live music is always better music.
You have 2 classics the grunge version and the acoustic guitar version
Thanks!
Hi Friends, I collect and play electronic music. When I search tracks to download, I reject 99 and keep 1 out of a 100. With AI this painfull procedure will increase to 1 out of 1000....
That is an argument of the record labels. AI music will over saturate the market. I’m pretty sure that happened once songs could be recorded in bedrooms and distributed with a few mouse clicks. But I can see how this compounds the problem. I still want to believe the real gems will rise to the top, regardless of if they are organic or synthetic creations.
It usually takes me about three days to write and produce a finished song with udio.
That includes concepting out the lyrics, usually writing some kind of chord progression I feed into the machine, drafting with udio on both the lyrics and performance side of things, and listening to a hundred variations of the song around 1500 times... usually in sections. And lately, in order to get that sweet sweet studio sound, I'll usually remix and bake my songs past ultra a few times. Everybody has their process, and it's all valid. That's mine
That is something I’m learning. Everyone has a different process and it is all valid. There seem to be so many workflows, the tech is still so new and still evolving. The workflow you describe is closer to what I normally do with Udio. The mass production of 50 songs was more of a “hit create, hope for the best and keep moving” experiment. I’m gonna have to try your technique of remix/ultra output.
@@ThaebrymMedia It sounds really amazing. So far, on my channel, the last two or three songs have been made with a process like this.
And I've got another one dropping later today (Dance With Me) on my channel that was made with a keyboard lick and me singing in the shower. But you would never know that by listening to it.
I think that's my new process. Crude midi renders and shower singing, lol. And then giving it to udio to clean up and produce. Then baking it to death.
Like you said, it moves so fast. The capabilities of the machine are moving so quickly that it's hard to nail down something persistent.
That's the big reason I've been afraid to make videos on it.
@@LynnColeMusic “Crude MIDI Renders & Shower Singing” would be an excellent album title!
@@ThaebrymMedia yes!!!
I've been generating songs in Suno and remixing in Udio (then EQ and editing in DAW) to mostly good results.I hate the 2 min 10 second limit cause I gotta remix the 2nd part of the song and join the 2 parts - the sound doesn't always match up well. For some reason Udio V1.5 is generating music that sounds dated and old while Suno is excellent at modern pop songs with strong melody lines. BTW I checked out your SoundCloud - good stuff! The sound is very polished.
I don't know English, so I read subtitles automatically translated into Polish. However, I got the impression that you were disappointed with your results. Don’t worry that someone claims to be able to generate 50 tracks in a month. Music is not a race. And the question is whether someone presenting those 50 tracks would have the courage to perform alongside the legends of the genre they create. Let me describe to you my working style with UDIO and the effects.
Since April, I’ve been creating tracks that I’ll only publish later. They are not visible on UDIO. Generally, it's Nu-metal, death metal, and other forms of rock music. I have 1,200 generations at my disposal, and I create 4 to 7 tracks per month, each lasting 4 to 8 minutes.
I use UDIO 1.5 and generate 32-second fragments, which gives me many options. One track can take hundreds of generations. I write the lyrics myself. I’ve noticed that when UDIO generates a track, it pays attention to the lyrics, to what the song is about, and sometimes it reflects this in the distribution of vocal positions and sometimes in the number of vocalists. I can't quite describe it, but the program seems to consider this during its mastering process. I've also noticed that when working on a single track, if you do too many generations, UDIO gets tired and starts making mistakes. That's when I set the work aside for the next day and start on a different track.
The only method for a great track is generating, generating, and evaluating, then selecting the best pieces. Currently, I have 10 rock tracks that I wouldn't be ashamed to present in the company of some top-tier metal bands. I mean those in the first league. Generating 2-minute fragments is a waste of chances to discover gems. You have to aim for quality, not quantity.
@@Pusty159 I agree 100%.
50 tracks in a day would fry my brain. I've been doing ten per day recently, and I think it's been too much. I'm going to take a break from Udio and come back when there's been a new update that hopefully takes away some of the frustrations.
I was interested to see your 60-30-20 method. For me, making grunge/garage rock music, I've found the best results from 17-0-0, but that's with automated lyrics. The lyrics can often come out garbled, but I find it produces more interesting melodies and variances. I have to put the lyrics back to 100 when adding instrumental parts or it starts adding gibberish. I never have clarity above 0, it just seems to make it worse to me, though perhaps I need to be more experimental.
I enjoyed your three songs, I'd pick number three as top of the pops myself.
You can't think too much about it. Just find some setting that work, change the tags every now and then, spam the Create button until you hit 50 hope for the best. V1.5 takes longer than v1 to generate something I'm happy with so this was a fun little experiment but nothing I'd do again.
mono.... 2024
I've done some AI/audience/listener experiments both in person and on my YT channel. Here are my findings.
Most people (including musicians) cannot differentiate AI music from the real thing when they aren't expecting it.
If they are told before listening that it's AI music they will pretend it was obvious and say they don't like it.
When the truth about the AI music is revealed they will be completely confused and refuse to believe it.
After enough exposure to a well collated selection of AI tracks most people seem to accept it as an interesting curiosity that often produces worthwhile music.
My conclusion is that what matters to most listeners is the result, no matter how it was derived. If you like the track, that's all that matters.
[Pretty soon even those of us on the alert, with our listening skills heightened, won't be able to tell the origin of what we are listening to.]
I agree, the majority of folks will not care who/what made it. The only thing that will matter is if they like it. And yes, these models get better everyday. Thanks for sharing your insights!
SUNO is better overall.
I think there is a great pop sensibility baked into Suno regardless of the genre it’s creating in.
People are saying AI music is going to destroy musicians. I beg to differ. I stopped making music due to a death in the family. AI allowed me to feel music again. I'm now on 150 ready tracks deep and the hardest part in production now is sorting tracks 😅 so that's 3 albums and I can't call them all BS now can I? So now I have shown you both quality and quantity 😉 SUNO, LUMA, DALL-E 3, and some python code. ruclips.net/video/v9eWYKkO8uY/видео.html
I use Udio as a band, calling it The Digital Bunnies, while I play electric guitar and write the lyrics, like I did in my song "I Am in a Dark Place
ruclips.net/video/sNkNZxW-G2U/видео.htmlsi=dkYKceCKXodwQlB1