SDXL has been out for a while, and there were people in the group who liked it. While waiting for the big shot's SDXL, I trained one myself and after just one day, the big shot released an update for SDXL. I didn't want to throw away what I trained, so I trained for another two days, but the results were only average. SDXL已经出来有一段时间了，群里也有人喜欢它。大佬的苗SDXL很久没更新，在等待大佬的SDXL的时候，我自己训练了一个，仅仅训练开始一天后，大佬就发布了苗SDXL的更新。我不想把训练的东西扔掉，所以又训练了两天。I have limited skills, so the characters are not very high-definition. Before training, I searched online and found that there are over 200 types of authentic Miao ethnic costumes, with silver accessories being a representative symbol for Miao women. 我水平有限。培训前，我在网上查了一下，正宗的苗族服饰有200多种，其中银饰是苗族女性的代表性标志。The traditional costumes are not overly revealing, but this model mainly focuses on travel photography and portraits, rather than traditional Miao costumes. So I added a few images of shoulder-baring outfits to the training set, incorporating or showcasing elements of Miao culture or traditional aesthetics. These costumes combine the style, patterns, or themes of traditional Miao costumes with modern designs or materials. They are specially designed for cultural or artistic performances to highlight Miao cultural elements. 传统服饰并不过分暴露，但该模式主要侧重于旅行摄影和人像，而不是传统的苗族服饰。所以我在训练集中添加了一些露肩服装的图像，融入或展示苗族文化或传统美学的元素。这些服饰将传统苗族服饰的风格、图案或主题与现代设计或材料相结合。专为文化艺术表演而设计，突出苗族文化元素。I tested the model on various popular models such as base official baseline, ZavyChromaXL, DreamChaser, Ronghua 3.0, and other large models. It works well with a default LORA weight of around 0.7. If you switch to a different model and the results are not good, you can increase the weight. I recommend running the model at a higher resolution of 2K for better high-definition restoration. 我在base官方baseline、ZavyChromaXL、追梦XL、容华3.0等各种主流底模上测试了该模型。它在默认 LORA 权重约为 0.7 的情况下运行良好。如果换用不同的底模效果不好，可以增加权重到1.0。我建议以更高分辨率 2K 运行模型，以获得更好的高清修复效果。Also, please don't choose the baseline based on its popularity with Lora 1.5. I have tested it and the results were not satisfactory. 另外，请不要根据 sd1.5 的喜好来选择底模。我测试过，结果并不令人满意。补充一下：提示词中增加skirt,dress搭配不同颜色tag，换衣服颜色；增加headdress头饰等tag添加帽子等。I would like to express my gratitude to DukeG for their technical support and guidance, and to Chirenshuomeng for their inspiration and example images. 我要感谢 DukeG 的技术支持和指导，感谢痴人说梦的灵感和示例图片。

苗元素衣服SDXL|Hmong Elements costume

SDXL has been out for a while, and there were people in the group who liked it. While waiting for the big shot's SDXL, I trained one myself and after just one day, the big shot released an update for SDXL. I didn't want to throw away what I trained, so I trained for another two days, but the results were only average. SDXL已经出来有一段时间了，小组里也有人喜欢它。在等待大佬的SDXL的时候，我自己训练了一个，仅仅一天后，大佬就发布了SDXL的更新。我不想把训练的东西扔掉，所以又训练了两天，但结果只是一般。I have limited skills, so the characters are not very high-definition. Before training, I searched online and found that there are over 200 types of authentic Miao ethnic costumes, with silver accessories being a representative symbol for Miao women. 我水平有限，所以人物的清晰度不是很高。培训前，我在网上查了一下，正宗的苗族服饰有200多种，其中银饰是苗族女性的代表性标志。The traditional costumes are not overly revealing, but this model mainly focuses on travel photography and portraits, rather than traditional Miao costumes. So I added a few images of shoulder-baring outfits to the training set, incorporating or showcasing elements of Miao culture or traditional aesthetics. These costumes combine the style, patterns, or themes of traditional Miao costumes with modern designs or materials. They are specially designed for cultural or artistic performances to highlight Miao cultural elements. 传统服饰并不过分暴露，但该模式主要侧重于旅行摄影和人像，而不是传统的苗族服饰。所以我在训练集中添加了一些露肩服装的图像，融入或展示苗族文化或传统美学的元素。这些服饰将传统苗族服饰的风格、图案或主题与现代设计或材料相结合。专为文化艺术表演而设计，突出苗族文化元素。I tested the model on various popular models such as base official baseline, ZavyChromaXL, DreamChaser, Ronghua 3.0, and other large models. It works well with a default LORA weight of around 0.7. If you switch to a different model and the results are not good, you can increase the weight. I recommend running the model at a higher resolution of 2K for better high-definition restoration. 我在base官方baseline、ZavyChromaXL、DreamChaser、荣华3.0等各种主流机型上测试了该模型。它在默认 LORA 权重约为 0.7 的情况下运行良好。如果换用不同的型号效果不好，可以增加重量。我建议以更高分辨率 2K 运行模型，以获得更好的高清还原效果。Also, please don't choose the baseline based on its popularity with Lora 1.5. I have tested it and the results were not satisfactory. 另外，请不要根据 Lora 1.5 的受欢迎程度来选择基线。我测试过，结果并不令人满意。I would like to express my gratitude to DukeG for their technical support and guidance, and to Chirenshuomeng for their inspiration and example images. 我要感谢 DukeG 的技术支持和指导，感谢池人硕萌的灵感和示例图片。

Files with SHA256: 723eed32869f14982416bdbcab85a50e12d39084fd68b5c04ac543ac9e3b0a9a