Firedotinc style.
Anima info: Anima version, should just work with the latest comfy or forge.
Newbie info: Newbie 0.1 lora version.
Description
Anima version, should just work with the latest comfy
FAQ
Comments (14)
Anima made me come back to local, and the lora works great. GOAT
True, model is really great
@bakariso For sure, are you planning to do your AIO style lora on anima? Wonder how that would work out.
@SkibidiGeorgeDroyd it's actually already being trained, because I'm curious to look at how it goes too, but I won't go for a full dataset for it, just ~15 or so
@bakariso Right on, if you upload i'll test it.
@SkibidiGeorgeDroyd it's not even a halfway through right now, so it would need some time to finish and be usable
@bakariso Got it, hope it turns out good.
What kind of training device was used?
4090 with 48gb of vram
@bakariso Modified one I assume? How do I get my hands on one by chance in the US if possible? Thinking about saving up money and getting a new GPU that would be great
@GPUPoorChad Yes, modified one. You could just use almost any gpu with 12+ vram with anima, it's absolutely not mandatory to train with huge batch and will fit perfectly
@bakariso I run LLMs as well definitely more beneficial there
Thank you for the Lora!
I used your worklflow. Did you also notice that the upscaled image had lower saturation of colors, or is it just me?
Yeah, that workflow is far from optimal, model output artifacts if you set resolution higher than 1024, so you would need tiled upscaling methods to prevent that https://civitai.com/images/120067460 like in this workflow