I saw a notice in old trainer for ACE-Step audio LoRA training and figured I"d give it a try. I was unimpressed with preview samples and had other priorities to publish first so I didn't publish it right away. Immediately upon attempting to do so I can conclude this feature didn't even receive any testing before release because there is a major, showstopper class, bug that renders it 100% impossible to publish anything without breaking ToS or using an external program with heavy learning curve. There's also a second bug that's impossible to miss (manual correction, doable in UI, is needed to proceed) but is easily fixed. I'm not going to spoil either for you (I've reported both, and it will be clear when it goes live in 4 hours and 20 minutes), but I'm impressed at how obvious the error was.
Edit: Impacted model is up.