For this plug-in, I suspect generating the best glitchiness would come from deliberately applying lossy compression, then attempting the conversion. See the section here for an example of what to do:
https://github.com/haoheliu/versatile_audio_super_resolution?tab=readme-ov-file#-why-do-cutoff-patterns-have-such-a-huge-impact-on-audiosr
Otherwise all the excellent advances in ML upscaling will probably result in a sound that is perceptually clear and broad-spectrum, and perhaps not very interesting
