AnimabaseLora in2.9b？
I apologize for bothering you during your busy schedule, but I would like to ask if the LOA trained on Animabase can provide a normal user experience on your 2.9b version, and whether it is necessary to retrain LOA?
Hi there, it will require complete retraining. I've tried expanded the lora itself without full retraining and it does run, just not very well. I will provide Lora Training scripts shortly.
Okay, then could you please tell me if this is based on training the base model?
Okay, then could you please tell me if this is based on training the base model?
Yes, Anima 2.9B was based on anima-base.
Hi there, it will require complete retraining. I've tried expanded the lora itself without full retraining and it does run, just not very well. I will provide Lora Training scripts shortly.
Okay, regarding LoRa, I'd like to ask if if we attach LoRa to the original 28 layers instead of expanding it, would we avoid retraining?
It may run, but it won't perform very well at all, expect melted textures, weird details or coloring.
Is it possible to blend styles with this finetune?
Is it possible to blend styles with this finetune?
I do use several artists for blending, it does work and many of the images on the page are made from blending artists, but its limitations are still similar to regular anima.
I've created a project to convert Anima Base weights to Anima 2.9B, which can significantly improve LORA's performance. You might want to give it a try: https://github.com/storyAura/Anima2.9B-Lora-weight-conversion