There are a lot of Calm's style LoRAs for Pony. I tried them all, mostly of them are bad or hard for me to get stable results, especially with LoRAs of other characters. So I made my own. This is not to mean my LoRA is the best one; that just means, in my cases, by my opinion, it works better than other ones.v1.1: Use strength 0.9-0.6, 'CalmStyle', and '3d, render, photo, cgi' in negative.v1.0: Use strength 0.8, 'CalmStyle', and '3d, render, photo, cgi' in negative; that works in the best way in most cases. Higher than 0.8, a character will have colored hair most of the time, so you might need to increase the color of hair you want. With LoRAs of real persons (and probably with characters with realistic datasets), the same goes, and strength should be 0.7-0.6; lower than 0.6, it starts getting 3D despite the negative, but feel free to try. Addetailer/only masked inpainting requaired. <s>I found an issue with succubus'/demon's teeth (mostly, but not only), they get orcish-style fangs</s>. Not really, actually it is usual behaious of vanilla Pony. Use 'sharp teeth' in negative to suppress it.

Calm's style

I turned back to the settings I usually use for Lora training, erased about a quarter of the images from the dataset, remade the captions so that I paid not enough attention in the first version. This version has better teeth, eyes, is a bit easier to use and combine with other Loras.

Files with SHA256: 93b0ebb23c3f552c4ea7c088c221320c4bbb536ba94a18899b71fa46c9e9ed39