sluo … that’s the co-CEO and founder… they’ve been using deep-learning since years ago, for ZivaRT--if not before that at Weta... if you get what I'm saying. They've already done it the SUPER SUPER SUPER SMART AND EFFICIENT way, lol.
It's just a suggestion for us for our circumstances and the progress we're at, if I'm not mistaken.
Like, it's probably better than what I would've just said, which is "start over" and set everything up in a way where you don't need two meshes; and you take into account the "settling". But that can be a big hassle too, starting over.
But I've been able to achieve the deformation youre after without using goal attachments. So you dont even have to do it that way; you can do it pretty much the ol’fashioned way; but it can be difficult and time-consuming in-and-of-itself doing it this way. But up to you. Pros and cons with a lot of things. We gotta' make do with what we got.
ADRS's may be quicker in the long-run, compared to the amount of simulations in trial-and-error I conduct.
Still fine-tuning it and not perfect, but many ways to skin a cat.