Open Toora opened 10 months ago
Oh, so it wasn't just me... rendering was going for like 5-6 hours instead of 2-4min as usual.
try using a smaller input image and see if the problem is still there
Was rendering done in minutes? Trying it in Summer 2024, [LamaLargeInpainter] Inpainting resolution: 1512x2048 with input image [DefaultDetector] Detection resolution: 1280x1536 . Takes ~5 hours no matter on what storage and which GPU.
Issue
I updated from a way older version (March perhaps), and noticed the rendering didn't seem to use GPU at all and was taking minutes instead of seconds. Working my way back through the commits it started with the "Improve hyphenation algorithm" commit 58a722a, on Sep 7. Before and after that is night and day. I used the same options.
python -m manga_translator -v --mode batch --font-size-minimum=17 --use-cuda --translator=sugoi -l ENG -i H:\T\
My GPU is a 4060 TI, CUDA and PyTorch using 11.8.
Command Line Arguments
No response
Console logs
No response