So this is a model that I combined from a few others surrounding the photorealistic premise. In the beginning it was just me trying to get a better photorealistic model, but I ended up finding that it has a special niche that I think it does really well in the realistic category.
If you are training your own lora of a character, person, or celebrity, this model tends to accept the lora into it very well. The model you train it with matters a bit, but I was able to get better results with this model when using loras trained on a specific person than the other popular realistic models out there.
It can definitely take a lot of tweaking and fine tuning when using a lora with a model so give it some practice and hopefully you will get some results.
Description
First Version, probably last.
FAQ
Comments (3)
Sorry for the newb question, as I don't know the first thing about training, as per your explanation. But does it also mean that this model is good at accepting the poses from other Loras, and will bend easily according to what a Lora asks of it?
No worries, none of this is really an exact science, it takes hours of training to get a pretty random outcome and I just noticed that this model was fairly flexible when it comes to loras. A blank canvas if you will.
I like the idea of this, and I've seen that it outputs well, but I am having to trim some data for space. Just wondering if any chance to get pruned? I got kohya to start messing with training models, but really no idea about it yet. Tried to convert to fp16, but got a missing module error that I will have to figure out. Let me know, thanks!