Hey everyone.
I'm a visually impaired person who's really gotten into AI Art. This is going to be the first merged checkpoint I'm posting.
This is a merge of two really great SDXL models.
Hentai ProtoGenesis XL by Dark_Infinity
and
Better Than Words v30 by waterdrinker
The main purpose is to make very NSFW images, and it's pretty good at it. It works well with simple natural language prompts with enhancements using typical tag based prompting. As such this is an incredibly horny model.
It also mixes well with many different types of LORAs, TIs, etc.
···················································
--=[PONY UPDATE]=--
···················································
After a fair bit of messing around I think I've managed to create an interesting hybrid mix of SDXL and Pony geared towards realistic results.
This is a mix of a number of Pony and SDXL Models.
This is using the following SDXL models.
https://civarchive.com/models/430454?modelVersionId=479579
https://civarchive.com/models/258840?modelVersionId=291876
It's also using the following Pony models
https://civarchive.com/models/341631/sevenof9ponyrealmix?modelVersionId=465600
https://civarchive.com/models/358760?modelVersionId=400990
https://civarchive.com/models/291042/hghdtri-shaper-xl-v10
Since this is a hybrid model it can use both SDXL and Pony LORAS (you might need to bump certain LORAS above 1.0).
You can use Pony's score_up nomenclature to change how the images are rendered. You can also use a lot of SDXL photography and lighting terminology.
This model can produce fairly good results at stock SDXL resolutions. Highres fix works well as does adetailer for faces.
··········································································
--=[PONY UPDATE THE SECOND]=--
···········································································
Here are a few interesting finds that should help with generating a wide range of images.
IRL, Fantasy, Photo, Unreal Engine 5
These four can control the look and feel of an image quite dramatically and can be used together for various effects. I find IRL has the most dramatic change.
score_9_up,score_9,score_8_up,score_8,score_7_up,score_7
These can change the results quite significantly too. Play around with each single one, or multiple together. Anything below score_7 I found to be of lesser quality.
skin blemishes,skin imperfections,skin pores,imperfect,
These together can improve the quality of the “skin texture” and other “realistic” things.
Speaking of realistic, don’t use it. I find it degrades image quality. YMMV 😊
Description
Initial Release
FAQ
Comments (19)
I like the mix of real and hentai... But could you provide more info what trigger words are possible?
There are no real trigger words that I'm aware of.
I do recommend "SDXL Styles" found here https://github.com/Douleb/SDXL-A1111-Styles
It allows you the dramatically change how an image is created. I think there is also a file for ComfyUI. If you use Fooocus it's already built in.
For photo realistic stuff you want to add in things like photorealistic, Hyperdetailed Photography, lighting types, etc.
Use negative prompts to remove things you don't want. Notice my examples basically have no negative prompting at all.
@jensgruber8659 since the model's recipe includes my Hentai ProtoGenesis XL model, it should trigger specific acts and poses based on natural language, but positioning keywords do help (like "cowgirl", "doggystyle", etc.).
@Dark_infinity Thanks for that amazing model. It's incredibly good for the type of stuff I like to make ;)
@stanleypain007313 try mix his other models as well. My fav is suzannes xl and deep dark 4.0
@Omikonz Technically some of those are already in here since the base model is the same across all of them. At least as far as I remember. I am working on something with my two mixes that involves PonyXL. It's tough going though.
@stanleypain007313 try using my pony hybrid to mix instead of the base pony. https://civitai.com/models/385945/deep-dark-pony-turbo-scoreless-sdxl-pony-hybrid?modelVersionId=430695
@Dark_infinity I'll give it a shot! Thanks :)
PONY models were trained heavily with the Danbooru (or a similar) dataset, so often you'll find the keywords here work as described, give them a try - they're very flexible and the reason I pretty much only use pony models anymore:
https://danbooru.donmai.us/wiki_pages/tag_groups
@Gouda Yup, there was also a lot of hand tagging done across a very large dataset. I find that using a small sentence to "set up the scene" at the beginning of the prompt + tags works very well here.
@StanleyPain I just posted something that may work well with this model. I’m using these wild card prompts with my Pony hybrid. I made them by using an AI to tag an ungodly number of images and then combined them all together as prompt wildcards. It’s here if you want to try it out with your model: https://civitai.com/models/460604?modelVersionId=512615
@Dark_infinity I've already downloaded those lol. Was going to take a look at them momentarily. Thanks for the heads up ;)
Please feel free to share your creations. I'm really curious to see what others can create. <3
Does unREAL Pony Mk. II contain VAE? Because only with SDXL VAE I get normal (not reddish) skin tone. After using Adetailer skin became red.
@StanleyPain I' ve solved the problem, My Forge works great with SDXL VAE (as you can see in my new gallery). Thank you for your work and such a good checkpoint!
Checked none vae in SwarmUI (comfy backend) - none (internal) vae produces overcontrast result (shadows became darker, details lost) and reddish result - skin is red. With SDXL VAE all colors are right.
Make this available for on site generation.
@hunta876 It is enabled.