使用50张1024*1024分辨率，带有wlop强烈风格的图片进行了8000步训练，数据集使用gemini3精细打标。使用8步加速的情况下，建议Lora强度1.50，采样器euler，调度器simple。wlop画风复杂难以模仿，建议多次生图以获取最佳效果。与flux版本相比单次生图成功率，画面美学，风格相似度都有所提升。Lora没有设置提示词，但测试发现某些特定提示词内加入“手绘风格”“厚涂”“笔触”这类关键词效果会更好。<h3 id="trained-for-8000-steps-using-50-images-at-10241024-resolution-all-in-strong-wlop-style.">Trained for 8000 steps using 50 images at 1024×1024 resolution, all in strong WLOP style.</h3>The dataset was carefully captioned using Gemini 3 with high-quality tagging.In the case of using the 8-step accelerationRecommended LoRA strength: 1.50Sampler: EulerScheduler: SimpleWLOP's painting style is very complex and difficult to imitate perfectly. It is suggested to generate multiple images to find the best results.Compared to the Flux version, this LoRA shows clear improvement in single-generation success rate, overall aesthetic quality, and style similarity.No trigger words were set for this LoRA, but testing has found that adding keywords such as "hand-drawn style", "thick paint", "brush strokes", etc. in the prompt often yields significantly better results.

Files with SHA256: 519da5d63d8acff4fb14a005d08ad2632f23c69ab002a14ea4a890698a038c1b