Description
FAQ
Comments (10)
What are your thoughts on sdxl? I feel like I might wait till more people make edits of it to before starting to training with it.
I thought SDXL would benefit more with full body shots for the dataset instead of close up shots.
(when it comes to anime, since it has less details to begin with)
Maybe it would improve better posing, proportions, and can grab more details of the character?
Looking at the current SDXL checkpoints and LORA's, most anime images that are closeups doesn't seem to have as much details
Good to know, I've not been able to get my kohya to handle it lol. I am only able to run it on comfy ui. My graphics card isn't to bad probably got to mess with some settings.
it much better listen prompt, better hands, it have better base quality(in fact for 1.5 anime upscaller is necessary), but it need much better hardware to train correctly
@Doomdom what videocard you have?
using a 3060 atm
@Doomdom i think you can make LoRA in kohya_ss with cache encoder and "optimizer_args": "--network_train_unet_only" this one, (ofc you need to finetune all other settings cuz 12GB is minimal for train LoRA)
you gotta make ayaka sunohara, would look fire
sorry, not in plan to make SDXL LoRAs, and not in plan to make ayaka sunohara.
guys, why you use this LoRA?
it SDXL one, and its bad SDXL one.
P.S use PonyXL one with Pony models(autismmix is pony too) https://civitai.com/models/406820/fatekaleid-liner-prisma-illya-illya-kuro-miyu-pdxl-and-sd15
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.