Do you want to support me? Consider being a patron on Patreon to vote on train polls and access stuff early. ❤️
For my 200th upload I wanted to recreate my custom mix from 1.5 MurkySpecial Mix .
As before this model heavy NSFW based
This model is quite experimental , I merged couple SDXL anime models then trained with 1024x1024 images from the 1.5 version then repeated couple times till it gives better results. I believe it captures the style of original versions bold lines and cute characters. Mostly for anime girls so sometimes 1boy prompt might not work, either prompt as "man" or try giving weight.
Discord server feel free to drop and say hi ^^
Description
FAQ
Comments (9)
If you trained it AFTER merging my good sir -- that's a trained model :P
It's on a grey zone really so I leave it up to community🤷
It's again on couple merges so I can't just say "Trained" model and ignore their efforts.
Yet also It's not just simple mergemix cause there is like 10+ hours trained data
Nice style. What models are included in the mix?
@Dark_infinity I trained the model seperately both on Animagine and ConfettiMix. Then combined them and added a bit more Style lora.
A bit of abomination but as long as it works 👀
@Murky, whatever works and turns out in the end, I've tried something similar (parallel development before merging). I've been a long-time fan of your 1.5 LoRAs, glad to see you doing more on XL!
@Dark_infinity Thanks, I wish it was working better for NSFW out of the box though 👀.
The only reason this is a checkpoint instead of lora is SDXL sucks at flexibility. Pose loras would just not work on most models at all -_-
@Murky Yeah, I know what you mean. Nothing is as easy or predictable as 1.5 was. But I'm having some real fun with the challenge of getting NSFW models to do what they are told. After a few months of work, I completed a few decent ones. But what SDXL really needs is a talented trainer with some good datasets that are curated and not watermarked to hell. You are welcome to merge, mix, or train on anything I've put up.
I love this model! Thanks for bringing it over, Murky :D
This does carry over some of 1.5's bad tendencies, like with hands. Even with that said, XL is just so much better at coherency that this is a huge win.
I really love this particular style, and I've had a hard time getting something close to it in SDXL til now.
No, or not much as I tested.
Trained the model couple times and generated new set with SDXL and trained again. Though it's AI, expect the unexpected 👀