You have to use this new checkpoint: https://civarchive.com/models/2855007/anima-29b
Description
A test lora for anima 2.9B. You have to use that model not the normal one
FAQ
Comments (6)
I first learned about the 2.9B version through your LoRA. How is the LoRA training going with 2.9B?
It went very nice. I just had to change the model and the repo on my script and left everything else the same. I have a int8 convrot version on my account btw
It's expected that you didn't heard about it before, the first day it was just a random post on reddit 2 days ago and there wasn't comfyui support. Now it's supported and it has lora training, everything happened very fast
@sovietlat I see, thank you! I looked into it and found out it's unofficial, but it's a huge-scale fine-tune, which reminds me of NoobAI. If this becomes mainstream too, local image generation will get even more exciting, so I'm looking forward to it.
@sovietlat Would you mind sharing what program you used for training the LoRA? If it's Anima-Standalone-Trainer, could you tell me what graphics card you have and what parameters you use? On my graphics card, a single training step takes 4 minutes, whereas on a standard model it only takes 10 seconds
@prodoge227173 I will send you a DM, I don't train with my gpu