Guidance needed for Full finetuning of SDXL

[Anustup900]

I am looking to full finetune the base SDXL model, not Dreambooth / LoRa approach.
So asking for best practices for those who had did this before for :

• what should be a good Learning rate to keep ?
• choice of scheduler
• choice of optimizers
• and steps needed per image ?
  Looking forward for some guidelines here

[asomoza]

The training scrips here are for learning and experimenting, probably not that much people trained a full SDXL model with the sample scripts, specially if you're going to use a big dataset for which the sample training scripts aren't optimized for.

If you want to do serious SDXL training, I recommend SimpleTuner which is based in diffusers and you can refer to the tutorial for some recommendations.

Even if you're not going to use it, you can use the codebase and the nice explanations to learn what you need.