No trigger word needed
Only makes faces
Weight of 1.0
Helps make faces more realistic
Good at making fictional characters real people
Handles prompting of ages, ethnicity, and physical attributes well
All samples were generated with Base SDXL 1.0. No refiner / detailers / highres fixes.
This LoRA was trained on over 100k high quality, highly labeled faces. It is just a small part of my Humans dataset. More information on that, and the thousands of tokens it has in it, can be found in the description of my Humans model. There are no trigger words and I do not recommend merging this into your model as it only does close up faces, unless that is what you are going for, in which case, go for it.
SDXL is amazing, but it is still lacking severely lacking in the ability to make photorealistic humans, especially faces. This was designed to help with that, but it is not perfect. Eyes and teeth are better, but still not at a level I am happy with, but I can only do so much with a LoRA.
I have also been training and tuning a full realistic SDXL model based on my full and expanded humans dataset since SDXL 1.0 was released, but it has a long way to go before I will be happy with it.
Description
No trigger word needed
FAQ
Comments (7)
wow. This looks so cool, I can't wait to try it.❤️
This does all ages, or only elderly?
It will prompt for younger. It probably has an older bias because the dataset is balanced amongst age groups and the dataset SD is trained on is not, so the mean age is higher, therefore making the bias higher when the two are compared. It it trained with terms like early/late 20s etc, so those terms should help with setting the desired age.
@Ostris thanks, balance is cool. So we're just seeing the average, that makes sense!
I guess average of 0-100 is 50. Hadn't thought about it that way.
Nightmare Fuel :D
Your models are really great. I would like to make a suggestion for a slider based on face shapes, I really don't know if it is possible. Like in this image: types of face format - Bing images and that one. Thank you!
