Possible to train LoRA?

black-forest-labs / flux

Official inference repo for FLUX.1 models

Apache License 2.0

15.48k stars 1.11k forks source link

Possible to train LoRA? #31

Open BlueZabu opened 2 months ago

BlueZabu commented 2 months ago

Hey, really loving the model. The team did great work. I saw a post on Reddit earlier that claimed fine tuning was impossible. I was wondering if the same applies to LoRA creation? Thanks!

oliverban commented 2 months ago

Everyone is patiently waiting for devs to answer! Some people say the released models aren't able to be finetuned or Lora trained but I don't know enough to make a sound analysis of it! We just have to wait and see what people find, FLUX has exploded so it will be quick to get answers - if not from the devs then from the community!

bghira commented 2 months ago

use simpletuner for training flux loras. needs 16G GPU.

monk-after-90s commented 1 month ago

https://github.com/monk-after-90s/ai-toolkit