This model is a mix of some of the best Pony v6 models fine tuned by different style LORAs. The goal of this model is a possibility to create beautiful images in a semirealistic style without the need of using additional LORAs.
As it is based on Pony v6 model You should expect the type of content it can create.
Pluses of using this model (in my opinion):
Ability to create beautiful semirealistic images only with standard prompt without the need for additional style LORAs (though you can still use them);
A decent anatomy and hands/feet/fingers (comparable to any other Pony V6 model).
Decent prompt adherence (mostly NSFW, but it can do SFW too)
Flexibility in different styles (to a certain degree, leaning more into semirealism)
Minuses:
Somewhat identical face type on generated images (without additional prompt).
Some color bleeding in images in purely 2d style (can be fixed by either stronger prompting or fiddling with seed/cfg)
UPD 2.1: An attempt to add "dirtier" aesthetics to the mix. The outputs look more grainy and have more contrast. Also I noticed more vivid expressions. The downside - a bit less accurate hands. Main component added - an excellent model by @phinjo: Ri-mix - PONY XL.
UPD 2.0: Back to more realistic roots (you can always use 1.2 if you prefer more cartoonish version). More clear facial features and improved lighting (shadows).
UPD 1.2: This time the changes are more drastic (introducing new model to the formula - CassyCartoon). This variation leans a bit further from realism, has higher saturation, darker tone and sharper edges.
UPD 1.1: Some fiddling with the formula of a mix, but no drastic changes (except updating to a 2.2 version of Pony Realism). The result is improved lighting and facial structure.
For my generations I usually use following settings:
Sampler: Euler a, DPM++ 3M SDE, DPM2 a, DPM++ 2S a (slow, but very detailed) - others work too, but those 4 are the best in my opinion.
Scheduler: Karras, Automatic, Exponentional, Restart
Steps: 25-100 (depends on sampler)
CFG: 4-10 (depends on complexity of the prompt and sampler)
Clip skip: 2
Hires fix: R-ESRGAN 4x+ Anime6B or 4x-ultramix balanced with 0.3-04 strength 10-15 steps
ADetailer: face_yolov8n with 0.2-0.3 strength
The models used in the mix:
The LORAs used in the mix:
https://civarchive.com/models/264290/styles-for-pony-diffusion-v6-xl-not-artists-styles
https://civarchive.com/models/470073/popyays-epic-fantasy-style-or-pony-and-sdxl
https://civarchive.com/models/142674/doujinshi-booba-sd15-pony
https://civarchive.com/models/341353/expressiveh-hentai-lora-style
https://civarchive.com/models/390981/realistic-anime-style-lora
All the credit and my gratitude goes to the original creators of these models and LORAs.
Description
Back to more realistic roots
Improved facial structure
Improved lighting (shadows)
Almost fixed color bleeding
FAQ
Comments (5)
Will you make this usable for civitai image generation?
I asked the support about this and they said, that there is a limit to the amount of models available for onsite generation and right now they can't add any more models...
@Axelros That's a shame, cannot figure out how to use any other methods of generation haha.
@MunkyDust There are some links in my profile to the places where you can use my models for generation. But I haven't updated all of them to 2.0 yet
excelente modelo , muy recomendable .
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.