UPDATE: https://civitai.com/articles/4121/sdxl-lora-training-guide-2024-feb-colab
New article for 2024 with colab link and video walkthrough :)
https://colab.research.google.com/github/Linaqruf/kohya-trainer/blob/main/kohya-LoRA-trainer-XL.ipynb
Linaqruf has a version available you can use to train these LORA:
LORA_LierLa
LoRA_C3Lier
DyLoRA_LierLa
DyLoRA_C3Lier
LoCon
IA3
LoKR
DyLoRA_Lycoris
Kohya's LoRA renamed to LoRA-LierLa and Kohya's LoCon renamed to LoRA-C3Lier, read official announcement.
I found that using the T4 GPU you can train batch 8 at 1024,1024 you mileage may very depending on the choices you make.
Batch 12 was under 24GB so i think batch 20 should run with A100 40GB
This colab requires you to upload your images to google drive and will output the built lora to your google drive after you allow it permission.
NOTE: You need your Huggingface Read Key to access the SDXL 0.9 repository, this is an official method, no funny business ;)
its easy to get one though, in your account settings, copy your read key from there