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
Use this only if you want to use artist tags or to train it further, without a lora in its way. Not recommened for normal use! Highly unstable in style, limbs and hands with high tendencies to gen realistic faces.
I have train it with a high learning rate latetly for testing and it got more unstable but better with artist tags. Because normal 1.1 is inteded for normal use, i picket this one with better artist tag recocnition, but is essentialy worse in every part in compare to the lora merge that was already in its base model more stable.
FAQ
Comments (13)
@Ikena, do you need help with datasets/captions?
no, but thanks for asking :)
I'm delighted to see that you released a model based on Pony. I had the opportunity to test several Pony based checkpoints and the renderings are very promising, particularly with regard to the hands, feet and details. etc
Is it possible to obtain results similar to your Sudachi model from this Hassaku XL? By playing with the flat_color prompt for example? Or do you foresee an evolution of Sudachi?
If you know a particular artist with a style you like, you can try it on 1.1, 1.1nonLora works better for artist tags and changed styles like flat_color, but its not great with details like hands and limbs and such. Im not home currently but i don't think that flat_color works great.
About sudachi, I actually done some lora tests to force a look like that of sudachi, but results where all highly unstable. Maybe because base is already highly trained to not make "flat" looking images. I put my attempts down, because i need to prepare for sd3 for now.
@Ikena
Hi Ikena. Thank you for your feedback. Correct me if I'm wrong. I haven't followed the developments over the last 4 months, but if I'm not talking nonsense, your previous Hassaku XL checkpoints are based on SD XL and your latest Hassaku XL checkpoint is based on Pony Diffusion XL. Which Pony XL itself is based on SD XL Lighting? And from now on, you would like to focus on the SD 3, also called Stable Cascade?
I'm gradually trying to move away from SD 1.5 because I realized how quickly SD XL worked and was much more effective at avoiding distortion.
The flat_color tag works very lightly with Hassaku XL 1.1, but you have to add a lot of strength to the tag and raise the CFG scale to 9. Unfortunately I can't get close to the rendering of your Sudachi SD 1.5 checkpoint which I really liked. It's cool if you manage to develop a loRA or even a port based on Pony. But I understand that you have other priorities.
Finally, I confirm that the v1.1 no loramerg model does not allow you to obtain very good results with the hands and feet. He even tends to try to avoid representing them.
@MangousteAI
your previous Hassaku XL checkpoints are based on SD XL and your latest Hassaku XL checkpoint is based on Pony Diffusion XL. Which Pony XL itself is based on SD XL Lighting?
--> hassaku XL is based on SDXL and was in the middle merged with animagine. I see that model as not done and a single 4090 can't train a good model in time. pony and animagine was trained on 3 A100 40gb in compare. This hentai version have as base pony. Ponys base model is also SDXL and not lighning. It is much more trained to the point that it can be seen as its own model, that why pony got his own category here on civit.
you would like to focus on the SD 3, also called Stable Cascade?
--> stable cascade is stable cascade, its own thing. SD3 is a not released yet new model from SAI. It is the successor for SD2 and would have 3 sized model to choose from on release. One of those models have the same size as SD1.5
Hope that answers all your questions
@Ikena
Thank you for your answers. I see things more clearly with your explanations. A lot of things have evolved over the last 6 months on the Stable Diffusion side and I was a little lost I must admit. I understand better the enthusiasm for Pony XL which saves a lot of time.
Good model.
Pony = I want delete my minecraft world irl
This model = new minecraft world
Amazing as always.
Does this need a VAE?
no vae is needed, vae is baked into the model
I was gone for a while and im happy to see you already made a new model! Can't wait to try it out
This model gets a update next monday, to make it more stable in style and limbs :)
This is probably the best model that I have seen on here, however, the model being inconsistent is stil a prominent issue. I am aware that this is able to create stunning images, but the quality varies too greatly to be of much merit.
Good job nonetheless.
Btw, I have tried both 1.1 versions.