Downloads · 30 days
28
0% of all-time downloads
shindi/realistic-skin-style
realistic-skin-style is a text-to-image model from shindi. Use it when you need an image from a text prompt. It is set up for diffusers. The card lists the license as creativeml-openrail-m.
You run your new concept via diffusers Colab Notebook for Inference. Don't forget to use the concept prompts!
Downloads · 30 days
28
0% of all-time downloads
All-time downloads
6K
Public
Repo size
7.7 GB
Likes
40
Public
Click a slice to open those files.
.bin5.2 GB · 67%
From the Hugging Face model README
You run your new concept via diffusers Colab Notebook for Inference. Don't forget to use the concept prompts!
This is my (Saleh) experiment with training on detailed photos of people featured textured skin. I noticed typically in my generations, skin looks airbrushed and smoonth which gives away that it's AI. Hoping to use this as a base model for further fine tunes so that models can look more realistic.
🚨 This is V1 of this exploration. Ideas for further exploration is annotating the training data as described here: https://www.reddit.com/r/StableDiffusion/comments/zcr644/make_better_dreambooth_style_models_by_using/
Note that you can use parenthesis (like this) to increase attention on a word or phrase. For example "A woman with ((blue eyes)) wearing a (((Hijab))), jqkz". More parentheses means more attention.
"A smiling black woman, jqkz"
"A white woman wearing a white shirt, jqkz style"
"A photograph of an old man wearing a black shirt, jqkz style"
"A young man with a cigarette, jqkz"
"A woman with ((blue eyes)) wearing a (((Hijab))), jqkz"
