This Version of Hassaku is the hentai variant of my other hassaku xl model. But again, like my 1.5 model with a priority for hentai art. It is trained on top of pony. Goal is to make it easier to archive good hentai anime art, without the need to use complicated source and long quality prompts.
My Discord for everything related to anime models and art.
If you'd like to support my work, you can do so at: SubscribeStar, every bit helps and is truly appreciated!
____________________________________________________________
Supporters:
Thanks to my supporters tamashiicolle, pttcot, SETI and Kodokuna
_____________________________________________________
Using the model:
Use this extension to use it easier with booru tags:
https://github.com/DominikDoom/a1111-sd-webui-tagcomplete
Use score_9 in positive and score_4 in negative
masterpiece -> score_9
best quality -> score_8
high quality -> score_7
medium quality -> score_6
normal quality -> score_5
low quality -> score_4
worst quality -> score_4
You can also use score_8_up, score_7_up and such, but effects are reduced.
Model is merged with a style lora, to make its style stable and to higher up the quality. Because of that, score_4 and score_5 have less effect, Lora /Checkpoint was trained with score_9, but can also be leaved out, for a bit more basic look.width and height should be a multiple of 64, to prevent gray borders (not a model issue, its a general thing for sdxl models)
Tag order: 1girl/1boy, character name, everything else in any order
Some concepts appear with strong 2.5d faces. Negatives like nose, lips and sources helps (in 1.1 reduced and in 1.2 rare, in 1.3 not a problem anymore)
It is a continuation of pony, source_pony, source_furry, source_cartoon, 3d, monochrome, realistic in negative works. Even with less effects on it
Model is trained with images with minimal disruptive features like floating texts, logos, speak bubbles and signatures as possible. If such features appearing in a image, please use prompt "signature" as negative.
Excluded metadata and franchise tags, don't use them. meta data -> something like "highres", franchise tags like "re:zero kara hajimeru isekai seikatsu" are pruned on characters like "rem_(re:zero)". Pony was trained on them, it seems.
Here are some good resolution options for SDXL/Pony:
1024 x 1024
1152 x 896
896 x 1152
1216 x 832
832 x 1216
1344 x 768
768 x 1344
1536 x 640
640 x 1536
License info:
Because this model is a continuation of pony, from PurpleSmartAI, it have this license:
This model is licensed under a modified Fair AI Public License 1.0-SD (https://freedevproject.org/faipl-1.0-sd/) license.
The following modifications have been added to Fair AI Public License:
You are not permitted to run inference of this model on websites or applications allowing any form of monetization (paid inference, faster tiers, etc.). This applies to any derivative models or model merges.
Description
Style is more anime like on some concepts and is overall more bright
What is better:
much more stable in style and limbs
slightly better artist tag understanding (nonLora 1.1 remains much better at it)
it is a strait up update of mostly everything (in detail and style)
What is worse:
Backgrounds changed, feels more illogical, but is much better at certain backgrounds in compare to pony
Can be to bright for some concepts
Faces appear less "mature" and having a more generic anime look, for some, thats a negative
Problems:
Overall too bright, working on that, except if SD3 comes out. So bright in fact, that dark night images are hard to get.
Eyes are sometimes undefined (blurry or circle pupil or such)
Some strange image artifacts can appear
FAQ
Comments (16)
score_4 is not worst quality. Scores can be used only in positive. Score_9 - score_4 are levels of high quality.
I train it like animagine and use the score prompts that are already there in pony as basis. Score_4 is indeed worst quality for my model.
The 1.2 artifacting is a damage to the unet. Use of denoised, upscaled, or generated images in training is not recommended.
can you let it run on civitai generations?
very good model, can draw good hands and feets.
although can't understand kemonomimi other than canines and cats.
also I generate pregnant girls, and maybe because there's a lot naked breast in training image, the belly button also sometimes transformed into nipple
It can definitely generate other kemonomimi. It helps to prompt it both with the "[creature]girl" prompt and the specific features you want. I tried it with a mouse girl and an elephant girl and it got the shape of the ears and tail right (I think I'd have to play with it a bit more if I wanted to get it to stick a trunk on her face but that's a bit too far into furry territory for me to mess with).
@person012345 what I meant is other than canine and cat girls, the model always generate extra human ears
@PurplefinNeptuna Ah. Well I imagine that can be prompted away? Actually in my test generations it didn't do that anyway, even with hairstyles that wouldn't hide the ears. Doing it now, mouse girl has mouse ears only. When I remove the mouse girl related prompts it goes back to generating human ears. Perhaps it depends on the animal and the situation.
This is honestly too much of a pain in the ass to use. It's VERY picky about how you phrase things in prompts, and simple things like adding weights using ( ) or scores randomly breaks the entire thing and generates noise.
It has the potential to generate good looking images, but it's just too much of a fight getting them. There are much better models that are far easier to work with that produce the same quality with less fuss.
like what
that () is auto1111 and forge weight bug, that happens to every model
you don't "phrase things", you use booru tags delimited by comma and space
@rockfallseveryonedies628 no fucking shit. I'm saying certain tags that work on other generators just fine don't work on this.
@rockfallseveryonedies628, I stopped using natural language long ago. Tags and weights work really well. You just need to be aware that overweight-ing a tag may suppress other and some tags will not work together. But that usually affects all checkpoints. This one looks actually pretty resistant.