Indulge in your Digital Desire
Goal of this model is great creative NSFW.
I present to you a mature version of my suit of models. While the current approach seems to use pony models, I find those models to be frustrating to prompt for and are limited in the merge options.
Model is obviously very horny but is also capable of beautiful SFW imagery. Try it out!
Do not hesitate to share your creations, I'm curious about what people can create with the model! :D
For V1.X
I've been using various samplers with typically around 40 steps to good result.
Model is quite versatile on style, from full on anime to photorealistic.
To get photorealism, it helps to add (toon, anime) to negative prompt
Inversely, adding toon or anime to prompt helps get those styles
Versions
V2.2
Added some amount of Real Dream SDXL Realism Refiner V2
V2.1
Merged in the following 2 models for increased NSFW realism
Monolith6 (Or at least I think that's what the name was, can't find it here anymore)
Merged in my own model to increase creativity
Tales of Digital Dream
And some others I cannot remember, been cooking for a while.
For this one, expect increased realism while still quite capable at creative images, semirealistic images or even anime. I typically use DPM++ 2M SDE Karras with UltraSharp upscaler.
V1.2.1
Merged in the following 2 models
AfroditeXL ❗ Nude People Photography
V1.1
Merged in my own model Digital Life as well as an experimental version of Digital Lust. Also added some amount of Cunnilingus XL.
Model leans more toward photorealistic than V1.0.
V1.0
This model uses Digital Dream as the base, to which I've added the following models:
Deep Dark Hentai Mix - NSFW Anime
I also merged in a lora I extracted from Acorn Is Boning XL
Description
Initial version, see description
FAQ
Comments (4)
Nice model! Which version of Deep Dark did you use? I've been tweaking the formula for 4.0 and you seemed to have done well with getting a merge stable. Knowing the method/block weights would help me out in a quest for version 5.
Unfortunately, the Deep Dark merge was the first one I did and didn't keep track of the weight, but typically I use somewhere between 20% and 30%, and I'm pretty sure it was version 4.0.
Merge was done with auto1111 as a simple weight sum. For the Lora (Acorn is boning XL), I couldn't get the merge to work, so I extracted a Lora out of it using kohya and then, also with kohya, merged in that Lora at a 30% weight if I remember right. Hassaku was merged in at 20% (auto1111).
I played around with iniverseMix too but didn't like the result.
Also, goes without saying, thanks for your hard work!
@TalesfromOurDigitalLives, that's good enough data. I'm not 100% with the text encoder in version 4.0 -- it follows prompts well, but something in it is broken with a few associations. It happens sometimes when I try to train a model with concepts it already knows but conceptually generates them differently (like perspectives/POV's). I look forward to testing this model to see if the merge repaired the issue. The images generated so far look promising.
thanks for your work. a really good checkpoint!