i feel like i've seen this comparison made before, but LLMs, when used, are best applied like autotune. 99% of vocal recordings released on major (and even indie) labels have some degree of autotune applied. when done correctly, you can't tell (unless you're a grizzled engineer who can hear 1dB of compression or slight EQ changes). it's only when it's cranked up or used lazily that it can detract from the overall product.