模型是从大佬的waiNSFWIllustrious_v140练来的，大佬的模型是融合的，我这也填个融合吧。新人啥也不懂，随便练的第一个丹。原本冲着数据集的图片来整ai的，没找着合适的模型，干脆自己练一个了。学个一知半解直接开练，练完后发现其实想练的是画风lora，结果练成了提示词触发画风的模型了.不管了，反正能用，先用着了，传上来当个备份也好。正面提示词: masterpiece,best quality,amazing quality,huanxiangxiang,FantasyHome,负面提示词: bad quality,worst quality,worst detail,sketch,censor,采样器: Euler aCFG: 4.5分辨率大小应该与1024x1024像素差不多就行。竖图效果要比横图好一些，横图如果分辨率拉到手机比例(9:20)的话容易出现分图,9:16就会好很多。放大设置与大佬的waiNSFWIllustrious_v140设置差不多。如果画面光线泛黄可以在负面提示词中加入yellow light,orange light。最好在正面提示词中写明画面的色彩，只在负面提示词中加入的话画面还是会偏暖色系。懒得整可以直接复制封面图中的正面提示词。The model was trained based on the expert's waiNSFWIllustrious_v140. Since the expert's model is a merged version, I'll just call mine a merged one too.I'm a beginner and didn't know much at first—this is my very first attempt. Initially, I got into AI for the images in the dataset but couldn't find a suitable model, so I decided to train one myself. With only a basic understanding, I jumped right in. After finishing the training, I realized I actually wanted a style-based LoRA model, but ended up creating a prompt-triggered style model instead.Anyway, it works, so I'll use it for now. Uploading it here also serves as a backup.Positive prompts: masterpiece, best quality, amazing quality, huanxiangxiang, FantasyHomeNegative prompts: bad quality, worst quality, worst detail, sketch, censorSampler: Euler aCFG: 4.5Resolution should be roughly around 1024x1024 pixels.Portrait images tend to perform better than landscape ones. If stretching the resolution to fit phone screen ratios like 9:20 in landscape, image splitting may occur; 9:16 works much better.Upscaling settings are similar to those of the expert's waiNSFWIllustrious_v140.If the image appears too yellowish or warm-toned, you can add "yellow light", "orange light" to the negative prompt.It's best to clearly specify the desired color tone in the positive prompt. Just adding it to the negative prompt won't fully prevent warm tones. If you're feeling lazy, you can directly copy the positive prompt from the cover image.

Files with SHA256: a35454b25b644944725fdf600afff191dc38df33890328008649228d51d72c34