模型简介: AnimeXL-xuebiMIX的最新版本,相比之前版本大幅度地改善了图片的光影,让图片层次感和氛围感更强 另外本模型已经融合了LCM-LORA,无需再额外调用,因此使用本模型时推荐以下设置: 1. 采样算法选择Euler-a(根据琥珀大佬的意见,实测比LCM采样效果好很多) 2. 迭代步数6-10即可产出高质量图片,大大加快出图速度 3. 高分修复算法可自行选择尝试,迭代步数8-10为佳,重绘幅度根据算法不同自行尝试 4. CFG Scale设置3-4为佳 当然,这模型也并非完美,目前主要情况是肢体细节仍然不太理想,正在研究改进,希望大家能多多测试返图,谢谢 ~~~~~~~~~~~~~~~~~~~~~~~o(* ̄︶ ̄*)o~~~~~~~~~~~~~~~~~~~~~~~ 测试提示词和参数: Prompt: ultra-detailed,(best quality),((masterpiece)),(highres),original,extremely detailed 8K wallpaper,(an extremely delicate and beautiful), anime, \\, BREAK 1girl,orange shoes,solo,sitting,sky,clouds,outdoors,black hair,bird,upward view,blue sky,white socks,daytime,orange jacket,building,long sleeves,leaves,long hair,stairs,red headband,pump Rope,headband,bangs,cloudy sky,from_below,wide_shot, Negative prompt: cover eyes,(bad anatomy:1.5),bad hands,missing fingers,extra digit,fewer digits,3D, Steps: 10, VAE: sdxl_vae_fp16fix.safetensors, ENSD: 31337, Size: 768x1152, Seed: 209465676, Model: AnimeXL-xuebiMIX-V6(LCM), Version: v1.6.0-400-gf0f100e6, Sampler: Euler a, VAE hash: 235745af8d, CFG scale: 3, Clip skip: 2, Model hash: c75ac9af77, Hires steps: 10, Hires upscale: 1.5, Hires upscaler: R-ESRGAN 4x+ Anime6B, Denoising strength: 0.6 |
2人表示很赞