Five clips looking like chewing gum, with shorter script lines fixing most of it, suggests the mouth drift builds up over longer takes. Were the 5 bad ones the longest scripts, or did they have something else in common, like a certain kind of word or a fast stretch of delivery?

The claim that people leave the second the lip sync slips is a sharp one, because it puts the failure at one specific moment instead of blaming avatars in general. When you've seen that drop-off, how many seconds in did it happen, and was it a particular word or a pause that gave the clone away?

Building the identical follow up in both tools is a clean test, and the split you found, faster to set up in Make but cheaper at volume in n8n, shows up exactly once you added steps. Which steps did you add when the per run cost got ugly, and roughly how many leads a week were you running when it started to hurt?