LoRA/Dreambooth Sample images: Does sampling use LoRA?

[rafstahelin]

While training, I depend on sampling to get a sense of the knowledge learned.

It's weird that I dont see resemblance when I prompt for single character or even with the trigger tag.

Whereas, when I use the LoRA in webui, I get perfect resemblance.

AM I missing simething, or is the Sampling through Kohya not calling the LoRA itself, and merely gens from the base model?

Attached dataset image, kohya sample (the extra crisp), and the webui gen