Iris: all faces or none?

I restore Super8, and 8mm tape. I deinterlace with QTGMC in Hybrid and apply some denoise. I then bring the ProRes MOV into TVAI. When the Super8 film was scanned at 2K on a ScanStation, I find that Proteus and Artemis don’t improve things much. But Iris can help bring out detail in faces. But because these home videos were shot in a jumpy way, the sharp face reverts back noticeably to the original face. And often the other people in the video aren’t getting enhanced faces. For me, this is a deal breaker. Therefore, I wish there was a way to problem it to say, do all faces or do none.

maybe try and stabilize the video 1st with the stabilization AI?