← sources

I'm going to pull my hair out ! LORA Training

source: reddit (reddit:StableDiffusion)published 2026-10-14 02:00ingested 2026-10-15 02:40

retained text

Hi, I'm training a realistic Lora character on fal.ai. (With Krea 2)

I have 26 images for 1,000 steps, in 1024 format.

I generated my dataset using Nano Banana because it’s recommended and it works well !

But the the beard, hairy body, and everything resolution are blurry when I generate with the character LORA, whereas without the LORA they looked great ! I’ve tried everything why is this happening ?

document fetched 2026-10-15 00:35 · canonical https://reddit.com/r/StableDiffusion/comments/1x5f76o/im_going_to_pull_my_hair_out_lora_training