The H4 Real project aims to utilize different methods and models to ensure the lack of shiny

doll skin in an image and attempt to make a realistic looking photo. Much like a DLSR or

High end Phone.

Currently I'm running a XL and PONY project.

All Models are trained initially then merged (in the event it's beneficial) with my private

models I use as refiners and tunning tools. They've all been tuned to produce decent faces and

hands. A detailer doesn't hurt but wont always be necessary.

I will be updating this project as time goes on.

Adding more PONY and XL Models for different purposes. Perhaps a LoRa or two.

Original Model was trained on 2500 images none below 1024x1024, with the exception of 720p 768x1280.

I divided them as evenly as I could the images based on resolution and organized each dataset from lowest to highest.

I know thats mostly for me, however it took awhile so I count it as part of the labor >_<''. lol

After training it was refined and tuned to produce soft skin tones trying to avoid the shiny uncanny valley of the doll like generations you can get with some models. The recipe is lost to time again, I need to organize better.

However it consists of JuggernautXL HyperRealismXL PrunedModel256SHA357 All models were obtained on Civitai, or Huggingface. Primary method of merging used was learn difference. With some weight adjustments to JuggernautXL cause it has the most human range, unfortunately suffers from shiny skin.

All the best

H4

(b ' . ' )b [Approved]