Sign In

how to train a lora if there is only 1 image

0

how to train a lora if there is only 1 image

a075662e-c606-4534-82cc-6260e63aa75f.png

女主角 [侦探]-白银之城(Heroine [Female Detective]-Silver Palace) - IllustriousXL | Illustrious LoRA | Civitai

Original Chinese → English Translation

User (Cano250, 10 months ago):

Hello, may I ask if the dataset only uses official art and fan art?


Comments:

OP (King_Dong, 10 months ago):

No fan art, I only used one official art image and cropped different parts to train on.

OP (King_Dong, 10 months ago):

In other words, I only used one single image to train the model.


User (t1911640040940, 8 months ago):

@King_Dong ? A single image? Without using Kontext, won't the model overfit?

OP (King_Dong, 8 months ago):

@t1911640040940 No, just lower the weight and it's solved.

My attempt (random Witch from somali and the forest spirit ep8)

QQ_1773830427861.png

QQ_1773830348033.png

lora👇(not final version,hair is not stable)

QQ_1773830980737.png

0