Udio.com (AI music generation)

Hodders

Hodders

Hodders

Soldato
Joined
18 Oct 2002
Posts
5,244
Location
Riding my bike
Anyone played with this yet?

It was mentioned on a Rick Beato youtube video, so I had a play.

Within about 10 minutes I had this..... https://www.udio.com/songs/2QHKJA18bHsWTmaiquVVqV

Who needs "TayTay" anymore!

I gave it a prompt and it took it from there, I then added extended sections to the middle and end.

I'm quite frankly astonished! Mind blown.
 
Last edited:
It's okay, however, if you're listening on decent equipment it's noticeably bad in parts. The vocals are tinny and almost chipmunk like at times.

@mrk give it a go on your ananda nano's.
 
Last edited:
I was going to post, Rick Beato's latest video on the matter, however you've mentioned it. :p I thought the clips shared during the video were fantastic, but I couldn't tell which was AI-generated or not. The power of AI learning is pretty impressive.
 
Ayahuasca said:
It's okay, however, if you're listening on decent equipment it's noticeably bad in parts. The vocals are tinny and almost chipmunk like at times.

@mrk give it a go on your ananda nano's.
Click to expand...
True, but this is the "Model T Ford". Give it a couple of years and the potential is mad.

The generation quality is only set to "balanced" - be interesting to see the results at "high"
 
I've been giving Suno a whirl - sure the mastery/quality is lacking with a lot of mush and all the vocals have a very similar generic style, but it can't half whack out some tunes - next generation of this stuff will be very impressive. It is quite a bit better than other offerings I've tried for consistency over the tune as a whole - that element does feel like a human touch in most efforts.

Definitely a decent tool for placeholder music for things like game development, etc. and even good enough for incidental stuff if fleshing out say a big open game world with a more curated, but low time consumption, alternative to procedurally generated content.

It fairly easily made something which would fit a game like Quake 2 - with a bit of tweaking using the cover feature to fine tune it https://suno.com/song/8fb809ff-a538-4574-ab8f-234a37e52130

For LOLs I asked it to create a song about the Martyn Ware vs Rockstar disagreement with minimal prompts other than the style https://suno.com/song/d022a6b0-a03f-45f9-90e3-a2dc53e51bc0 it seemed to get the assignment if a little bit on the nose, with some quite cold disses in there hah though not sure they are entirely intentional.

It struggled a bit with covering a tune I made as a kid, some interesting reinterpretation of it but the quality took a noticeable dive, especially reinterpreting in a different music genre it really struggled. Most attempts were just a mess though it is a beta feature. Though this one was kind of interesting https://suno.com/song/7ac36f00-3365-460a-8845-12e720ea191a
 
They all sound bad to me, they are clearly generated by a computer, everything sounds compressed and weird.

unless you have good speakers you probably can't hear it.
Its like the AI has to much compression or not a full range of frequencies and can't properly replicate an authentic sound.


Everything should be sounding crystal clear on my speakers and it's not.

maybe the bitrate just ruins them or something, I guess they are heavily compressed and not even YouTube quality?
 
arknor said:
They all sound bad to me, they are clearly generated by a computer, everything sounds compressed and weird.

unless you have good speakers you probably can't hear it.
Its like the AI has to much compression or not a full range of frequencies and can't properly replicate an authentic sound.


Everything should be sounding crystal clear on my speakers and it's not.

maybe the bitrate just ruins them or something, I guess they are heavily compressed and not even YouTube quality?
Click to expand...

From playing with it a bit you get a range of quality, simple prompts will often produce something reasonably close to decent quality, more complex prompts or trying to create variations of stuff you like etc. can quickly turn into mush. Some genres seem to have high quality samples as well, while others sound like they were ripped from poor quality 8-bit sources.
 
Last edited:
V4 model for Suno released - the quality improvements mostly seem to be about tweaking the dynamic range to hide the mush, etc. which seems to alter the tone a bit in a not ideal way, though does seem to have structure improvements and vocals. The "remastering" of older generated content doesn't really improve things much IMO as again just seems to mess with the dynamic range rather than really improve things and/or the AI sees mush as valid input and tries to do something with it.
 
You must log in or register to reply here.
Back
Top Bottom