V1: version trained with old settings with unnecessary tag
V2: tagless version with better (I hope) settings that should perform better
Border in negative prompt could be also useful sometimes
Description
FAQ
Comments (21)
Awesome!!! Akifn style please !! I can share his uncensored fanbox torrent if needed! Its about 405 pics. And maybe abubu
Ty. Whats so special about their styles? Cant say its very appealing to me
@bakarisoAkifn has a really cool way of drawing genitals and skin, his style is kinda unique, and abubu is nice in anatomy. Heres akifn full uncensored fanbox gallery if wanna check https://files.catbox.moe/5z5crb.torrent
@Rating_Agent I'll check it once i will try fizrot
Do you have any guide how to make this loras? Would like to try to do some more!
Well, just use sd-scripts and copy the setting from the metadata of this lora ig if you have good enough gpu for that, but im still trying to experiment with them. There is already many guides on how to handle your dataset and train but the most interesting one for pony is here https://rentry.org/ponyxl_loras_n_stuff with a lot of styles too
@bakariso how to change the steps count in the script ? i have only 3100, probably its not enough
@koloved333 Which script are you talking about? Should be more than enough though. Its --max_train_steps arg for sd-scripts
any possibility for a 1.5 version?
There is already one made by someone https://mega.nz/folder/c3p1kZoI#N-dac0u42jF2Bx-cWN4Slg/folder/EmgkURAR or you can use easyfluff with hll, it recognized the artist too
@bakariso nice, where you foind the mega folder?
You got any tips for sdxl training? I've tried to do a few styles, but they keeping coming up pretty awful.
@bakariso Thanks!
PonyXL is Clip Skip 2, but it's written here that your Lora, made for Pony, needs Clip Skip... 1 ?
Initially, at release, there was no such information on the pony page, so yes, I use everything at clip 1
This lora is everything
Any chance you'll retrain your loras with Clip Skip 2? Or just retrain for general improvements? Not sure how much of a difference it would make though, these loras are the absolute best. Thank you for putting in the time to make them!
Clip skip is not a thing for XL models, I could retrain some old loras that permorms not that well with latest good settings, or to make tagless versions. Which of them you think could perform better?
@bakariso didn't know that about clip skip with XL thx for clarifying. Again I think they're great, but I'd say the Akifn one has shown the least stability along with worse hands/contorted bodies (at higher weights like 1). Not significant at all, just the one I'd have to single out if any.
@crazydude8450794 Well now, after the last updates, clip skip is actually doing something, but it's very insufficient and affects just the small clip_l, even if you skip all the layers there won't be that much difference between the pictures, but w/e. I can try to retrain akifn with better settings
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.