This is an extract of my Humans model. It took a while to find the right balance of size and fidelity, but I am finally happy with this version. No trigger words, weight of 1 works well. Scale up from 0 - 1 to add more skin detail and "ordinary person" features. Visit the parent model for more information, but the short version is:
This model is designed to produce photo realistic images of normal people. Most SD models can only produce beautiful people. This is not that. You will get acne, moles, ratty hair, crooked teeth, wrinkles, and well, ordinary people
There are thousands of trigger words which can be found at https://gist.github.com/jaretburkett/cf8c224243834172fc13f72aaf49811d , or for a sorted list based on frequency, see here https://gist.github.com/jaretburkett/41370fdf69b791d2b406f3fa538d4b32 . The big one to know is the word “face”. A significant portion of the dataset has faces, and they were all tagged with face. Use it to get faces, without it you will get farther away shots, usually portraits.
Description
Extracted using my toolkit script. https://github.com/ostris/ai-toolkit . Settings were:
{
"mode": "quantile",
"linear_quantile": 0.5,
"conv_quantile": 0.5
}FAQ
Comments (15)
May I ask a question? If I want 4 characters appear at the same pic, each with different features and expressions on their faces, how should I write the prompt?
I don't believe you can get that with any of the current stable diffusion or even midjourney models. It will take separated prompts per segments, and even then, not great.
Try Additional Networks loras or Composable Lora with latent couple, it should work.
Or you can use Tiled Diffusion & VAE plugin in webui
https://github.com/hako-mikan/sd-webui-regional-prompter.git
Tht is exactly what you need. look for a guide on YouTube. it's very easy
I know this is a lot to ask, but I wish this could also be a LoRA instead of a LyCORIS. I love it, no more perfect airbrushed same faces. This is exactly I have been looking for.
I wonder how this compares to the full model. Is the full model more flexible? Or are they the same?
I would probably recommend using the Lycoris with your favorite model. That will offer the best flexibility and alloy you to tone it up or down as you with. The extraction method uses allowed for very little loss in the way of results.
Everytime I try to use this, I get an error. "!is_dim_broadcast || non-broadcast_dim_size == 1"
Then Automatic 1111 DirectML crashes.
so no trigger words or there's thousands?
Not really trigger words. More of a custom captioning vocabulary. You do not need to use them necessarily, it works fine without them.
Not really trigger words. More of a custom captioning vocabulary. You do not need to use them necessarily, it works fine without them.
Can anyone tell me how to use LYCORIS file in comfyUI?
Did you figure it out yet?
It works EXACTLY the same as a lora. Just plug-in a Lora Loader Node.
You also put it inside Lora folder