OC Character Test, trained on base CyberRealistic Pony. To trigger: Lynn, heart-shaped pupils, V3 - I wasn't quite happy with the face in V2. To fix it, I generated a bunch of images using it on various realistic, anime and 2.5D models (including my own merge which many of the showcase images use, that I will upload). Retrained using 50 images for 2000 steps. My ComfyUI workflows should be available in my images. As before, use ADetailer or FaceDetailer to refine the face/eyes.

Lynn

V1 suffered quite a bit when attempting to use it outside of anime-style checkpoints. This time, she was trained on <a rel="ugc" href="https://civitai.com/models/4">CyberRealistic Pony</a> with realistic images, as well as some anime mixed in, in an attempt to A) negate some anime same-face, and B) increase flexibility. This version is also not overcooked like the previous, and a strength of 1 works perfectly well. LoRA works well on realistic models this time around (or at least cyberrealistic) as well as anime, like Hassaku.

Files with SHA256: 6e826858ae212c6aa16eed784d0241a465769df1f1283fd862c4457a327bb252