<h1 id="radiance">Radiance</h1>一款基于 MOSS-Image-SFT 训练的插画画风 LoRA，侧重细腻的动漫人物轮廓、发丝表现、冷暖明暗，以及衣料、金属和透明材质的局部高光。希望呈现的是有层次的光影与细节主次，而不只是增加泛光或压暗画面。<h2>使用建议</h2><ul><li>推荐底模： MOSS-Image-SFT</li><li>建议强度： 从 0.8 开始，根据画面在 0.8–1.5 之间调整。</li><li>普通采样参考起点： Euler，50 steps，CFG 3。</li><li>加速 LoRA： 推荐先关闭额外的八步加速／蒸馏 LoRA。</li></ul>目前测试中，不叠加加速 LoRA 时，画风呈现更好。如果使用加速组合，建议单独比较线条、上色和材质表现，再决定是否保留。欢迎分享生成结果和使用参数，也欢迎反馈不同提示词、强度和采样设置下的表现。<hr />An illustration style LoRA trained on MOSS-Image-SFT, focusing on delicate anime contours and hair, warm–cool lighting contrast, and selective highlights on fabric, metal, and transparent materials.The aim is layered lighting and a clear balance of detail, rather than simply adding more glow or darker shadows.<h2 id="recommended-use">Recommended use</h2><ul><li>Base model: MOSS-Image-SFT</li><li>LoRA strength: Start at 0.8, then adjust within 0.8–1.5 to taste.</li><li>Standard sampling starting point: Euler, 50 steps, CFG 3.</li><li>Acceleration adapters: Start without an additional 8-step acceleration or distillation LoRA.</li></ul>Current testing shows better style rendering without the additional acceleration LoRA. If you use an accelerated setup, compare the linework, coloring, and material details separately.Feel free to share your results, generation settings, and feedback!

Files with SHA256: 26965262fc6c72eadf6800fc59ebbdd5ba2709b1a0b1fc0e318d0e100ad11f10