Since AI Music platforms like Udio and Suno have been getting a lot of attention lately, I wanted to get some Hexbears’ opinions on the matter. Have any of you been testing the capabilities of these? Care to share what you’ve made?

Udio seems to be trained on a wider range of niche genres, which I think leads to more diverse sounds that are better at obscuring their AI origins. Suno has a much more limited mainstream range, but you can make an entire concept album by extending clips multiple times before you ‘get the whole song’. You can also finely tune each clip by choosing to extend it very early into the clip to pick out the best parts (though at that point, why not just make the music yourself?)

So far I’ve been using Udio to get more diverse samples, combined with an AI music splitter to isolate the good parts. I then plug them into Suno-generated long tracks to augment limitations from the prompting process. The music still isn’t great (and my editing skills are dirt poor) but I think interesting things can get created this way. Soundcloud link to some example tracks

Adam Neely just released a decent video about AI music and what it lacks, and so even though the critique of capitalist art production that he suggests is pretty milquetoast, I’d recommend it as a mild antidote to all the AI music hype.

  • dualmindblade [he/him]@hexbear.net
    link
    fedilink
    English
    arrow-up
    4
    ·
    edit-2
    4 months ago

    I’ve used them both, both are pretty impressive but I am much more impressed with Udio even though it has some limitations which suno does not. Basically it’s more flexible and creative, both are bad at mixing/defying genres but Udio is better, the sound quality is better, it’s easier to get something that sounds like a particular artist (for better or worse), but most importantly Udio is much much better at producing things that sound good outside of 4-5 popular genres. On the downside, it’s not as good at adhering to lyrics and it is impossible to get compositional coherence over 1 minute, I’m fairly sure this is built into the architecture, the “context window” is between 30 and 60 seconds, with Suno it isn’t great at coherence but it will remember things that happened more than 60 sec ago. So if you want something with verse/chorus type structure, or you want intermittent vocals and the same voice the whole way through you’re out of luck. Anyway, here’s some of the more interesting stuff I made:

    Only track that I consider entirely good, modern classical acapella chamber choir, nonsense lyrics: https://www.udio.com/songs/r6zuQYrhZoUhG2BPjJfkZ6

    Piano and female vocalist, lyrics by a family member: https://www.udio.com/songs/vZin6X5CFsbU8CXhwhRMGC

    A mix of genres that ends up turning into DubStep, lyrics by chatgpt and based on navy seal copypasta. In my opinion this isn’t good music but dear god does it go hard in spots. I’d rather listen to an album of this than Skrillex: https://www.udio.com/songs/i7xtbbgnsGNkriS27BYkx3