Second LoRA's is here! Trained on Chinese furry author: "A bowl of fish rice" 一碗鱼饭丶(<a target="_blank" rel="ugc" href="https://x.com/Fish7163">https://x.com/Fish7163</a>)<h3 id="please-support-the-real-artists!">Please support the real artists!</h3><hr />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: 5–8</li><li>Recommended steps: 24–44</li><li>Trigger word: f7163 </li></ul>For best results, please use the negative prompt examples from the OP – they significantly improve output quality.<hr /><h3 id="new-v2-version-is-out!">New v2 version is out!</h3>This version delivers significant improvements across the board due to long training:<ul><li>50 epochs total, improved tagging, and a cleaned dataset – no blur, less sketchy style, less bad eyes, only minor artifacts with some angles, and much more details. Supports a wide variety of furry species, and the results are now much closer to the original author's style.</li></ul>I hope you enjoy it, and please share your results! Feedback and reactions are always welcome – they help me improve future versions.<hr /><h3 id="resolution-limitations">Resolution limitations</h3>This LoRA was trained on images with a native resolution of approximately 1024px on the shorter side, primarily using a 2:3 aspect ratio. For best results, use a similar aspect ratio and resolution as from the examples from OP.Generating at resolutions below 1024px may result in:<ul><li>Blurriness and loss of fine details,</li><li>Distorted anatomy and structural issues,</li><li>Increased noise, artifacts, and overall degradation of image quality</li></ul>Recommendations:<ul><li>For best results, generate at 1024x1536 or higher (e.g., 1024x1920 (2:3) or 1920x1080 (3:2)), this matches the training resolution and aspect ratio of the used training dataset, delivering the best detail and stability.</li></ul>The default CivitAI generator and its built‑in workflow that using suboptimal settings (resolution handling) that hurt quality. It's better to use ComfyUI with the workflow provided below – it's based and fine‑tuned for this and other LoRA's.Example workflow used in ComfyUI:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/07ad3b7c-f130-4621-a33b-594a0412fee3/original=true/07ad3b7c-f130-4621-a33b-594a0412fee3.jpeg" />Be sure to check out the Suggested Resources tab too. Any tips, reviews, or feedback for future training would be appreciated!

一碗鱼饭丶Style

50 epochs, less artifacts, more to original author style

Files with SHA256: 071a48b74fe456899409d5bec896cff69418d7a665a0270d4119b5fa41ae0df5