Artist's resources:
https://www.pixiv.net/en/users/62713424
https://www.patreon.com/DittoChad (support the greatest :winky_face:)
Recommended samplers:
DPM++ 2S a - 20 steps
DPM SDE++ - 18-20 steps
DDPM - ~24 steps ( personal favorite)
Recommended upscale settings - here.
Recommended checkpoint:
NAI-XL vpred1.0 2d accelerated "Colorized" version
Illustrious version is here. Go check it out as well. Uhhh... Did I ever train one?..
Consider visiting or even joining my Telegram channel :3
https://boosty.to/mcnortrom - for lora commissions.
Description
I hope this is it
I hope this is it
I hope this is it
I hope this is it
I hope this is it
FAQ
Comments (3)
Does training on anynoob produces better results than training on noob v-pred?
Do you use both v-pred and ztsnr on the trainer?
According to my tests - yes, Loras trained on anynoob give better results. I don't use ZTSNR as it is recommended on Anynoob's civit page (you really better not to), but as far as I know if you are training on checkpoints based on V-prediction you should always check V parameterization. As for for "Scale V pred loss" checkbox (if you train your loras using Easy Scripts just like I do) - I didn't test it and not actually sure what that thing does. xd
Keep in mind that I am probably the most unreliable source of information regarding Lora training, so be careful asking me anything about it haha.
forget about it, anynoob messes with colors and shapes in terms of style training. just train your loras using base noob v-pred
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.