PDXL v1.0b
A retrain of 1.0 that's half the size, not overcooked, and more flexible. The only reason I'm calling it 1.0b and not 2.0 is because I didn't update the dataset like I eventually plan too and it also isn't as consistently accurate to the character as 1.0.
PDXL v1.0
It is probably worth waiting a bit for me to do an update. The model needs more training images so it had to go to the fifth epoch meaning it's a bit overbaked and requires a lower weight 0.7-0.8ish. Also, yes it's massive. I'm going to fix that problem with the next update. I didn't realize my settings were wrong until after the training.
If none of that bothers you and you just want to gen Tsukino then have at it and share what you create! If you make anything that's really good let me know and I might include it in the training because last time I checked when training the SD 1.5 version there was a serious lack of images.
Anyway that's all see you all again in the next post!
Old SD 1.5 Version
First off, yes, it can do her normal proportions. I'm just being safe with the image choices.
Second, the LoRA is marked as NSFW because it is hard to put clothes on her.
IMPORTANT TAGS (remove what you don't want):
anthro, blue eyes, leotard, brown body, brown fur, female, female anthro, fur, hair, neckerchief, paws, pink hair, stripes, tail, thick thighs, felid, felyne, lynian, mammal, tsukino \(monster hunter stories\), capcom, monster hunter, monster hunter stories, monster hunter stories 2 wings of ruin, skulkers
For Human Ver:
blue eyes, leotard, brown body, hair, neckerchief, pink hair, stripes, tail, thick thighs, tsukino \(monster hunter stories\), capcom, monster hunter, monster hunter stories, monster hunter stories 2 wings of ruin, skulkers
NOTES:
It really likes thick thighs so if you use the LoRA but prompt something other than the character with the 'thick thighs' tag and a lower LoRA weight you can get some good results.
As Always:
Trained on Anything v3
Demo is Anything v4.5 with LoRA weight 0.8
Description
FAQ
Comments (12)
It's great but I don't get the same as you, what model do I use? Since I'm using Anything v3 but when I use your prompts and your LORA I get the version of the game, I don't get that thicc version, help
I used anything v4.5 for all the demo images. Also make sure your LoRA weight is 0.8
It's a pretty rad model, based, thanks
add monster hunter or felyne tag, it helps find model
We realy need a Pony model of this lore
Well I just trained my first Pony-based LoRA so I'll work on this one soon. I'm going to retrain most if not all of my old LoRAs!
Done but expect a retrain in the near future
@thesauceer Holy speed m8, i posted this without hope, you're awesome, it is a little blury some times, but its preaty good for the first Pony lora, congrats
@otakugamerbr Yeah definitely a bit blurry. I highly recommend using it with adetailer to help fix that. Hopefully when I have a moment to look there are more images to train on which should help the blur factor. Enjoy!
Just a heads up I've done a bit more with PDXL and think the update I just pushed might be better for you. Hope you like it!
I just saw your gens! Very cute! I was looking at your settings have you tried using Euler A SGMUniform instead of DPM 2++ Karras? It might help with the artifacting around the eyes. I could be wrong though haha
@thesauceer Thx dood, Euler A SGMUniform ? did not tried that, need to make some reasearch because i only have the default ones yet, will try later, the update realy made some good progress, i realy liked