Udio AI Music Generator - Tips and Tricks
HTML-код
- Опубликовано: 6 окт 2024
- Just watch it on double speed, yeah?
#udiomusic #udio #aimusic
1:08 - a "Megan Thee Stallion"-like
2:07 - get genre labels from Wikipedia
3:14 - insert lyrics for better results
5:10 - trying for Enrique 1
5:40 - trying for Enrique 2
7:05 - the extend feature is hit or miss
7:49 - trying for Enrique 3
8:03 - trying for Enrique 4
8:30 - trying for Enrique 5
8:45 - trying for Enrique 6
9:16 - a "2NE1"-like
9:37 - a "Tech N9ne"-like
9:56 - trying for Enrique 7
10:10 - trying for Enrique 8
10:25 - trying for Enrique 9
10:52 - trying for Enrique 10
I've noticed putting a little bit of text saying what the song is about helps improve your results. I suppose it provides some context for the AI in terms of what emotions it's meant to convey etc. Udio is loads of fun!
I definitely need to keep playing! I think a lot of mood is pre-set when I pick stuff like 'kpop' or 'citypop'. Could be cool to try to make sad happy songs and vice versa though!
I have some great tips from what I've discovered: "19th century opera" often results in perfect and authentic sounding opera, which I've tried with both English and Swedish lyrics. Another often perfect and varied style is using the tags "indian, bollywood, sitar, hindi movie". The reason why it works so brilliantly I can only presume is because of how many Indians there are and how much musical content they produce, resulting in vast quantities of data for AI to train itself on.
Oh! Another great genre to try is "stand up comedy". You can accomplish some funny insanity with that. Especially if you manage to break the voices with weird screams and squeals. include some [laughs] and [scream] and [screams] at random places for fun.
(Bulbasaur always first pick for starter BTW)
Thanks for the tips! I made one symphonic metal song so far, and want to try to get with actual opera vocals like Nightwish. Let's rock 🤘
@@blackjackjonny Bulbasaur first pick... OH! No, no, no. Except for looking at cute Pokémons I didn't actually play it as a kid. Wasn't very big on it. For this account I picked the name Bulbasaur together with the word Brutal because I'm familiar with a forum trend of a community 8 or so years ago. On a white nationalist forum they'd sometimes run polls for all sorts of things and it became a meme to put Bulbasaur in it as one of the responses. And then Bulbasaur always won. If I recall correctly, one out of that community called himself Bulbasaur and was a member of a major white nationalist podcast that came out of that community, but he eventually had to be booted from it because it leaked that he was in fact a homosexual. Crazy story no? That's how Bulbasaur got so Brutal! Haha!
...
Onto the more relevant topic: Oh I feel you, bro. Nightwish has been on my mind a lot with composing AI music but it's not something I myself have attempted, but it was always in my mind because I've attempted to try to make metal songs in the style of Ghost BC. Mostly unsuccessful but some came spookily close! Would really like to be able to make a ton of jacked riffs backed up by Hammond synths while I wait for next Ghost album to drop.
Here’s a tip. Just have fun with it. If you come up with a good musical idea you can extend it and then eventually crop out the first part and you have the song you wanted
This is the way
Dude "namekian fury" sounds like Violent J so much it's scary! Even the beat is very ICP.
I think the 'horrorcore' label is a super strong influencer. The rest of the tags in my prompt were: angry, Hip hop, rap rock, hardcore hip hop, horrorcore, chopper, male vocalist-- in case you want to try it yourself!
@@blackjackjonny Thanks, I'll give that a go!
I work for an AI company and what's interesting is it's actually cheaper to pay people to train the AI than to put it online and pay for people to use it free. So basically people are doing work and not being paid since like in this video you are helping to train this software.
Thanks for that insight-- it makes a lot of sense! Cuz I feel like they have to be eating TONS of bandwidth with how fast it is, the "unlimited" storage, plus the image generation, AND audio/video exporting. I know huge upfront capital sacrifice is like techbiz101. Will be super interesting how it all plays out...
OpenAI totally did that with ChatGPT. When it first came out it was a lot better, then got dumber and more useless as soon as they first updated it. Originally it just warned you that it may go against T of S, but didn’t respect you AT ALL. Im sure this was for training purposes. GPT-4 is undoubtably way better without the limitations, which makes me think OpenAI has a way more powerful LLM in house which is the source of all the weird warnings and concern stories that came out a few months or so ago
10/10 lyricism
Can you add an accent prompt for the vocalist in addition to gender, American, English etc.?
I haven't tested this extensively, but I will say that if I do `k-pop` it'll have k-pop English. Then, when I do `japanese` I can hear the accent to varying degrees depending on what genre I associate (citypop, visual kei, japanese rock).
I think as long as there's an archive for it to pull from off Spotify (this is just my hypothesis), it should be able to do it-- and there's definitely plenty of k-pop and j-pop on there.
I think if there's a prominent type of accent for a certain genre, that might help influence it. I definitely hear some German when I try `europop`.
Hi so for some reason i’m getting a strange message at the bottom of the screen whenever I press create. Even when I do everything correctly, it says song is not ready yet. How long exactly does it take to generate? It seems like it’s also downloading something in the background. I am definitely on the correct site and I am definitely doing everything, correct, but I’m not sure what’s going on here. It’s very frustrating. Can you please try and help out? Thanks.
Hi Erick!
First thing I would do is submit a ticket at support.udio.com/hc/en-us/requests/new
Now that there's a price gate and a free daily credit, you definitely want to be compensated for things not rendering.
As for my observations: I've noticed a slowdown in how fast songs generate for sure. Maybe as of two weeks ago, and especially since they opened up the subscription model. I think the rumor is paid subscribers get placed on a priority queue. (I'm still on the free tier.) I want to say generations this past week took up to 2 minutes, but before responding to you, I went ahead and used my free dailies and timed them. What I got was 1 minute and 36 seconds, 1 minute and 22 seconds, 44 seconds, 1 minute and 4 seconds, 44 seconds.
I haven't seen any downloads unless I click to download the track. I didn't see any errors at the bottom, but I know sometimes during song generation, it will throw up an error message, almost like a placeholder as it's loading. (It will go away once the song has advanced in its generation.)
Hopefully Udio officials can get you all sorted and back to making some tunes. Good luck!
Aaaaaaaand NOW the ouput is limited, and has a subscription plan now.
Yeah, I wonder if they originally planned to wait a month on the dot, or if there was another reason...
Fake k pop stuff
trying for Enrique 12
www.udio.com/songs/5keoM5Dmtvbiiz9BroVV78
Pick up a DAW and make actual music
That too. This AI stuff is just junk food while the real food is still on the burner.
trying for Enrique 11
www.udio.com/songs/g6st5HGbepMEBfS6hx8oMP