The logs for the app (logs tab) state it's just repeatedly using the number 42 as a seed. But yes in the UI the results are different each run, even with an unchanged prompt. Didn't quite parse your last but if you're saying it's misleading, yeah kind of, because you can't share or save a prompt to redo the same result.
Weird that it's considered a big challenge to fix… something in Metal? Core ML? SIMD libraries? Some translation layer? Something is not being said here.