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
It is a merge with normal sdxl and my training to create a balance of nsfw capabilitys, character retain, look and backgrounds. I needed to do this, because to fix couple of things of ponys training, the most effective way is to merge models in, that are not "broken". But that also brings problems with it (see the last point).
What is better:
Contrast and colors are much toned down
Backgrounds
sfw knowlege
What is worse:
nsfw composition
nsfw precision (some nsfw prompts work less good then before)
Anime character precision
Artist tags
Problems that i want to fix the next couple days, because it is a merge with complete different unet, prompts ect:
Eyes have sometimes artifacts
Especialy "brown eyes" have problems
Genitals appears blurry sometimes (especially insertions)
Notice: censored in negatives censor or makes genitals blurry/white/black
FAQ
Comments (13)
Great model, I've been using it since version 1.0. I don't understand why people complain about the complexity of use. Everything I write works exactly as intended. However, I have always been interested in complex scenes and character interactions, which many models find challenging. I read the changelog for version 1.3 and it mentions that the quality of some NSFW prompts has been reduced, but personally, I haven't noticed any changes in that regard. Can you give examples of changes that you have noticed?
EDIT: many authors who focus on NSFW use the Pony model as a base, and if they don't, they eventually switch to it. Those who stick with the classic SDXL model, no matter how many versions they release, still can't compare to Pony in many criteria. What is the reason for this? Is it an insufficient dataset for training, or maybe incorrect methodologies?
Some prompts, like grabbing another's breasts and cunnilingus in POV view, don't work as well as before. If it's like that for those two, I can assume the same for other NSFW prompts.
As for why everyone switched to Pony, Pony's trainer claims to have three A100 40GB GPUs, similar to what Animagine's people have used. My other Hassaku SDXL model failed because my single 4090 is simply not enough to compete with that. It is simply a matter of GPU power and the time required to make models, and Pony's model is the only one that focuses heavily on NSFW. Everyone using base SDXL to train NSFW has a challenging time if they don't have the equipment for it. Anime models need even more training and care because of the many styles and the quality of images used in training. I have a dataset of 1.3 million images, but one epoch takes one and a half week. Animagine's people train the same amount in a month, but get 10 epochs in. Someone with the right equipment has it obviously easier to achieve results in a reasonable time. Pony's model has trained for three months on three A100 GPUs.
Are you plan transfer hassaku model to SD3?
I plan to train SD3 and make a new model, that would take a while
nice
Take my money... I mean buzz.
I've been fond of Hassaku since SD but was disappointed with the inital release of Pony. The style was unique, gorgeous, and singular in a plentiful of aspects I could go on about. The sore point of XL is that the overall style is nice but I have found Pony to be superior in terms of anatomy. This is a key variable in my eyes as I have lost spectacular runs only to an extra toe or hands that look like they came from some nightmare fuel.
Given this information, I was ecstasic at seeing Hassaku transition over to XL. Other anime models like Animagine (through Ponymagine,) have found themselves improve dramatically through becoming a Pony model. As said, I was elated at learning of the Pony version of Hassaku Xl. This was short lived, given the admittedly erratic artstyle of the first version(s). (Parenthesis are weird, so is English.)
You may have seen my previous comments on here that described my sadness at the performance of this model. I was still optimistic though, because I saw very solid generations every once in a while. Those expectations died with version 1.1.
Come version 1.2, I decided against caring much for the model. This changed with the release of version 1.3, as I remembered your statement of Hassaku XL development dying off with the release of SD 3. When that did eventually release, I finally gave version 1.3 a whirl. After tinkering a great deal, I concluded that it lived up to what I imagined the model could be.
Anyway, please take my money/buzz now.
Note: I haven't tried NSFW aside from the stuff the model gave me (no sex,) so I haven't mentioned that. Which isn't much of an issue for me because I love Ponymagine and Animagine. Both are phenomenal at what they do but not so for NSFW. I might try it later but for now I will continue repeating the same settings, prompt, and neg because I just love the results I am getting. Adios.
One last info I am leaving you with: Does the model know what a circus and lolipop are? I have tried generating that for a while. I'll most likely be posting some of those images, showing that it does not do as the prompt says. Other than that, it usually follows the prompt well. Kudos.
Something else I should add, I never actually tested version 1.2 until recently. I found similar results to 1.3, so I would most likely would have loved that version.
PS: If feasible, I would enjoy if further Hassaku models or anime models you release have similar outputs such as the ones this model generayes.
I have blue / blurry outputs.. and its incredibly slow to run. Any idea why ? I have a 3080 GTX and radeon 7 5000
Am I missing something in my stablediffusion perhaps ? @Ikena
PS Every 6gb +- checkpoint is going slow for me and with the same issues.
Anything between 2-4 gb checkpoints do work for me. Iff you know the answer lemme know pls !
2-4 are 1.5 models, this here is a sdxl model, do you use the right dimentions and don't force a sd1.5 vae on it?
@Ikena It seems youre right, I was forcing 1.5 with sdxl
I got rid of this problem when I switched to a different UI, not from 1111
@TheGata117 Yeah ive switched to ComfyUI which seems to be doing the job faster
Details
Files
hassakuXLPony_v13.safetensors
Mirrors
hassakuXLHentai_v13.safetensors
hassakuXLPony_v13.safetensors
hassakuXLHentai_v13.safetensors
hassakuXLPony_v13.safetensors
hassakuXLPony_v13.safetensors
hassakuXLPony_v13.safetensors
hassakuXLHentai_v13.safetensors
hassakuXLHentai_v13.safetensors
hassakuXLHentai_v13.safetensors
hassakuXLv13.safetensors
[Pony] Hassaku Hentai [V13].safetensors
Pony-hassakuXLHentai_v13.safetensors
hassakuXLHentai_v13.safetensors
hassakuXLHentai_v13.safetensors
hassaku-pony-Hentai_v13.safetensors
hassakuXLHentai_v13.safetensors
hassakuXLPony_v13.safetensors
hassakuXLHentai_v13.safetensors
hassakuXLHentai_v13.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.