This LoRA was produced using a new method. Started from a Dreambooth trained on Bender images, extracted as a large LoRA LoCON, Resized using sv_from 0.9. The resulting LoRA is very close to the output of the original DB.

Samples are not cherry picked and are straight from a basic prompt against the default SD1.5 base model. Face can be significantly improved using the LoRA + inpainting of the face.