This is a LoRA (1.5) for creating SPHINX for use in medieval rpgs.
Wow, these models with mixed animal/human parts can be tough and this was one of the harddest. Almost as hard as the Chimera I still haven't gotten back to finishing. I had some test models that had 30k steps and really looked terrible, so after refining the training images for a while and dropping the number of them to some solid different ones, I was pleased to get at least these results ;)
NOTE: The XL version came out much better IMHO, so I'm probably going to revisit this one to make a better model. The XL is just so much easier to get good images from. I would recommend using the XL until I get a better model for 1.5.
I would use contribution around 0.65 - 0.85, and there will still be plenty of misses.
Other than the main key rpgsphinx , it's probably worth having wings, claws, lion body, humanoid face in any prompt as well.
The trigger words for the model are the following:
all fours, black eyes, black fur, black wings, blue eyes, blurry, blurry background, brown eyes, brown fur, claws, closed mouth, dark skin, dark-skinned female, dark-skinned male, earrings, feathered wings, feathers, flying, full body, headband, headdress, horns, humanoid face, jewelry, lion body, lips, long hair, looking at viewer, multicolored hair, no humans, realistic, rpgsphinx, solo, white fur, wingsDescription
initial 1.5 version
FAQ
Comments (2)
This is very strange and uncanny. The face is quite different in treatment from the body, so it looks photoshopped.
I hear you and I understand.
there are stark differences at times, but are never quite "plopped" as in photoshop. There are some that have more gentle transitions which I'm rather fond of, even though they are still conceptually different.
The training images were not so abrupt, but getting the model to function in 1.5 was truly hard.
I'm building the XL model now and hopefully it will work a bit more smoothly.



















