自己尝试第一次炼的lora，训练集感觉不是很好找，所以这个lora可能会有些瑕疵，请见谅。
触发关键词：
aemeath,solo,1girl,
ahoge,multicolored hair,pink hair,ponytail,sidelocks,very long hair,hair ornament,heart,halo,symbol-shaped pupils,+ +,yellow eyes,
breasts,medium breasts,cleavage,detached sleeves,detached collar,leotard,gloves,covered navel,elbow gloves,strapless,white dress,white gloves,white leotard,
white thighhighs,thigh strap,white footwear,high heels,single thighhigh, boots
画质建议提示词：
line art,Detail enhancement,Melting,shadow,masterpiece,best quality,good quality,very aesthetic,A shot with tension,impactful picture,offcial art,movie perspective,masterpiece,facing the lens,best quality,very aesthetic,extremely delicate and beautiful,amazing quality,32K UHD,((solo_focus)),simple_background,sunlight,reflective,masterful details,intricate details,soothing tones,high contrast,natural skin texture,soft light,sharp,ultra-high resolution,32K UHD,dynamic_outline,Close-up details,outstanding style,adding a touch of dimension to your images without compromising details,iridescent textures,delicate details,magical atmosphere,concept art,highly detailed,ultra-high resolution,32K UHD,sharp focus,best-quality,masterpiece,unconventional supreme masterpiece,masterful details,with a high-end texture,in the style of fashion photography,Dynamic Angle,Dynamic posture,extreme aesthetic,masterpiece,best quality,good quality,newest,very aesthetic,absurdres,Visual impact,ultra-high resolution,32K UHD,sharp focus,best-quality,masterpiece,Emotionalization,unconventional supreme masterpiece,masterful details,with a high-end texture,in the style of fashion photography,(Visual impact:1.2),,impactful picture,offcial art,splash of color,movie perspective,masterpiece,best quality,amazing quality,very aesthetic,absurdres,best quality,newest,mature,incredibly absurdres,huge filesize,wallpaper,推荐的参数：
底模/checkpoint：oneObsessionBranch_v7Matureeps/waiIllustriousSDXL_v170
步数/Steps：28
CFG：5
权重/LoRA Weights：0.8（更换衣服时在0.6以下 Clothing variation weight set below 0.6）
采样算法/Sampler_name：Euler a(normal)
放大算法：R-ESRGAN 4x+ Anime6B
(p.s：建议使用R-ESRGAN 4x+ Anime6B，4x-UltraSharp放大反而会变得更糊）
降噪/denoise：1（高清放大，面部修复为0.5 High-resolution upscaling enabled, facial restoration set to 0.5.）
Description
生成的眼睛如果比较模糊的话，建议使用缩放latent放大来改善眼睛的模糊，再搭配面部修复，可以很大的改善此类问题（没有办法，这是sdxl模型的通病）。
也可以将图片直接拖入comfyui里面，使用我的工作流来生成，建议搭配fox大佬的崩坏三立绘lora来使用。
底模建议使用：oneObsessionBranch_v7Matureeps/waiIllustriousSDXL_v170
If the generated eyes are blurry, use latent upscaling to fix the blurriness, and pair it with facial restoration. This will greatly improve the problem. (This is an unavoidable common issue with the SDXL model.) You can also drag images directly into ComfyUI and use my workflow for generation. It is recommended to pair it with Fox’s Honkai Impact 3 character portrait LoRA.
Recommended checkpoint: oneObsessionBranch_v7Matureeps/waiIllustriousSDXL_v170
Comments (7)
LoRA 还原度很高，比其他人弄的爱弥斯lora像多了。但是感觉模型有点僵硬，3D风格比较强烈，而且随机性也比较大，生成不同的图片时变化较大，较难得到稳定的图片。
希望有办法画2D画，假如要爱漂活动好像身体怪怪的，模型有点适应不了其他姿势，希望有新的版本，感谢。