If I'm not wrong this change to workflow will ensure no matter the character lora/training data, the results will be similarly varied as this post for example.
Before, you may have noticed sometimes some characters keep getting generated in 1-2 poses and angles with small variations. (This only happened for characters with bad lora training/data)
But.. I think this improvement solves that entirely!
As a bonus it should help keeping the style more consistent as well.
DivinerSkye
2024-12-10 12:57:54 +0000 UTC