For best outputs: Version 11.0: Sampler: DPM++ 2M. Steps: 40, cfg 7, CLIP skip 2. VAE: none required. Pos prompt: this version listens way better than the previous one, being high weights rarely necessary. To create a furry make sure to add the word "anthro". It's not a trigger but the model will quickly understand that you want a furry with it. Quality words welcome, like "detailed fur" or "realistic". For example: "(anthro:1.3), female anthro horse, curvy body, front view, seductive, standing, overalls, detailed fur texture, ultrarealistic, detailed background, horse stable, outdoors, looking pleasured, messy, portrait, sensual pose, sensual smile". Neg prompt: "deformityv6, boring_e621_fluffyrock_v4, long neck, multiple tails, floating tail, decapitated, short legs, black body" plus what you don't want to see ("NSFW, less than 2 people" ect). Both embeddings, "deformityv6" and "boring_e621_fluffyrock_v4" can be found in my 11.0 version example runs. Version 9.5: Sampler: DPM++ 2M SDE Karras. Steps: 30, cfg 7, CLIP skip 2. VAE: clearvae_v2.3_fp16.safetensors. Pos prompt: add the LoRa's "add_detail" and "more_details" (both with a weight of 0.9) at the end of your pos (you can find them in my v9.5 example runs). Neg prompt: "aid210, boring_e621_v4, feral, long neck, multi tail, (bad anatomy:1.2), (bad hands:1.2), blurry, deformed hands, face out of frame, multiple tails, disfigured, deformed, malformed, mutant, blurry, extra limbs, extra fingers, missing limbs, amputee, malformed hands, multi balls, multi penis, floating penis, dialogue, text, deformed face, signature, logo, artist name, cropped, bad quality, worst quality" where the embeddings "aid210" and "boring_e621_v4" can be found in my v9.5 example runs. What changes from v9.5 to v11.0: input_image feature, overall quality and realism increased, better listening model, increased NSFW skills (check my v11.0 NSFW test runs). Checkpoint Merge.