What are you running, and what still fails?

[SysAdminDoc]

Two questions for anyone using this regularly.

First, the setup. GPU and driver, which OCR engine you settled on, and whether you're running CUDA, DirectML or CPU. The backend status panel in Help will tell you most of this if you're not sure.

Second, and honestly more interesting to me: the failures. I want to hear about the jobs that don't come out clean. Hardsubs over moving backgrounds, stylized fonts the detector misses, karaoke effects, subtitles overlapping station logos. If you can describe the source material or share a single frame, the failure pattern usually tells me which detection or inpaint path needs work next.

No format requirements, just reply. Screenshots welcome.