My first LoRA!Trained on Chinese furry author: "Luhua 5S Pressed Golden Germ Corn Egg" (卤花5s压榨金胚玉米蛋 on <a target="_blank" rel="ugc" href="https://weibo.com/u/5217510292">Weibo</a>)<h2 id="please-support-the-real-artists!">Please support the real artists!</h2> Based on <a target="_blank" rel="ugc" href="https://civarchive.com/models/1570986/yiff-in-hell?modelVersionId=3134413">Yiff in Hell </a>model.<ul><li>Recommended CFG scale: 6–9</li><li>Recommended steps: 40–44</li><li>Trigger word: 5schinese (more to original style, add these: spiky hair, small horns, dynamic pose, powerful stance, painterly style, vibrant colors) </li></ul><h2 id="new-v2-version-is-out!">New v2 version is out!</h2><ul><li>Fewer artifacts,</li><li>More training epochs,</li><li>More detailed faces and anatomy compared to v1</li></ul> The LoRA is still a work in progress – not perfect yet.Please take note negative prompt without these and similar keywords: bad quality, worst quality, low quality, worst detail, boring angle, bad angle, very displeasing — greatly influences the final generation results.Any tips, reviews, or feedback for further training are highly appreciated!Example workflow used in ComfyUI: <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/c26f47c9-70dd-4c01-bcdc-b23025046721/original=true/c26f47c9-70dd-4c01-bcdc-b23025046721.jpeg" />

卤花5s压榨金胚玉米蛋 Style

More references, less artifacts, 20 epochs, total ~30 images, and another approach in training

Files with SHA256: c63e411bfe0580faa2033abc6253b5cecd633537b48847b54a882774c4788665