Yeah, I actually use it quite a bit, just to fuck around with and make covers in differing styles. I think it's really lame that they're going to remove the 4.5 and 5 models. The way of the world I guess: just kill off something people enjoy in favor of making more money. I'm getting really sick of this type of shit. I read that the new models are going to be extremely "hard to fool", in the sense that you won't be able to disguise existing audio (which is what I do now) to get it uploaded and start working out covers. So, yeah, I guess I'll be waiting for something else to pop up. As you said, it's pretty much inevitable that something from China or Russia will appear sooner than later, possibly with those same exact 4.5 and 5 models. You know there's already a shitload of data-mining going on right now.
I just want some local versions to be released. There's a local version of Stable Diffusion and even ChatGPT, although you need a stupid amount of RAM for the bigger ChatGPT model (64GB is the minimum, I think). Just waiting for somebody to release something open source which gives us something Suno adjacent. Unlikely to be quite as polished, but hey, if it's free and open source then it's a lot harder for corporations to get their grubby hands on it and destroy everything. The irony of these record labels, who have been stealing from artists for decades with shitty agreements, pearl-clutching over AI is extreme. They only care because it might slightly dent their profits - they couldn't care less about the artists themselves.
I saw a post on the Suno reddit the other day, it mentioned a prompt for creating a "live" performance from one of the existing songs you'd made. To be honest I had never even thought about using it that way, but I tried it out and it actually worked really well. Some generations were admittedly better than others, and it wasn't perfect, but I got a half-decent "live" version of most of the stuff I'd completed. Pretty much ran all of my remaining credits out. The prompt, if you want to try it out, is:
"A high-energy live festival performance of this song, played on a massive outdoor stage in front of a huge crowd, Include loud audience ambience, cheering, clapping, and the crowd singing along loudly during the chorus, The crowd are singing along and allowed to sing some of lines, The lead vocal should sound live and energetic, with natural reverb and a raw concert feel, Instruments should feel amplified and powerful, like a real live show, with crowd reactions between lines and big audience participation in the hook, The overall sound should feel immersive, wide, and atmospheric - a full stadium or festival vibe"
The quality of the crowd noise and participation definitely varies. I'd say 1/4 of the time it's really good, and otherwise it'll either be passable or just so bad the generation has to be binned.
It's also worth putting an [Intro] tag at the start and putting something there, otherwise it tends to just have somebody saying random gibberish that vaguely sounds like a song introduction but is clearly meaningless. Giving it intros to use, I was actually shocked by how well it integrated the tag into the song. It tends to either add the intro over the top of the beginning of the song, if there's an instrumental intro, or the intro is your typical speak and then the song starts.
I'd say this was probably the best generation I got - and I have to admit, in many ways an outlier as none of the others managed to integrate the introduction quite this well and create what sounds very much like a proper live version of the song. Even down to the crowd singing in the background, which I didn't get in any of the others.
The only problem I found with the ones where the crowd are more involved, is that it tends to fuck up at some point. Usually by
just having the crowd sing certain parts, and unfortunately they usually sound insanely drunk while doing so. Usually it's the chorus, and if you have that repeated a few times then usually you can use one of the other choruses to fix the mangled one, but yeah... not perfect. Still very cool though.