Any way to avoid the typical "simple face, detailed body" look with anime models?
I've noticed that a lot of anime models (all the ones I've used except v3) tend to create images with a very distinctive look, where bodies have detailed and semi-realistic skin and fat, but the face is very simple and anime-like. That can be a nice style, but sometimes I'm looking for something closer to what anime-style art actually looks like. I've noticed that a lot of the time, when character LoRAs (or LoRAs for certain styles) are added, they'll make the art style of the whole image come out looking much like the original anime style. Is there any way to get that kind of look without using a LoRA to get a hyperspecific result?
For context, here's the kind of style I'm wanting to avoid (images taken from the Anythingv5 gallery): https://civitai.com/images/1379307
And here's an image (with a character LoRA) that seems to have a more consistent style throughout the whole image: https://civitai.com/images/1549504
You can try to make aesthetic embedding via aesthetic embeddings extension of Web UI with a images of a style you prefer. I myself planning to try this approach later to get my hands on a stable cartoon-anime style, which currently i can get only by using ControlNet.
There are quite a few style loras that can help with this.
Inazaki has several embeedings here that can give prettier faces / eyes etc https://civitai.com/models/33873
You can also use a lora like this one to make more "distinctive hot girls" https://civitai.com/models/96587
Honestly a lot of it is just using a better model though. Anythingv5 looks generic because it is essentially just the most basic baseline "ai anime image" you can make. I would recommend just using a different model that has better eyes and faces and an interesting style.
They will give you better faces. There's hundreds of fantastic models to pick from that have their own styles