时隔两年,又入了SD的坑。之前的图(Soultonic:Stable diffusion极致真实人像)虽然现在也还能看,但毕竟两年过去了,总得有点升级吧,于是就有了这篇。
UniPC
DPM++ 3M SDE
Checkpoint模型:xxmix9realisticsdxl_v10(c站上的)
**Lora模型:**之前用的胶片风lora的3.0版本(LEOSAM’s FilmGirl 胶片风 LoRA–全新VELVIA版本-LoRA-LEOSAM是只兔狲-LiblibAI)
正提示词:xxmixgirl,Thorough, analog style, eye focus, highest quality, (highly detailed skin), photo of a exquisitely beautiful pale skin punk Dutch girl, 21yo, (wearing harness, and beanie), perfect face, alluring eyes, [seductive makeup], skin pores, (piercing:0.5), indoor, messy bedroom, (bokeh:0.6), sharp focus, dappled lighting, (backlighting:0.7), film grain, photographed on a Sony A7R IV, 18mm F/1.7 cine lens, (highly detailed, intricately detailed), 8k, HDR, seductively posing, front view, (uppper body:0.9) <lora:LEOSAM’s FilmGirl 胶片风 LoRA–全新VELVIA版本_3.0:1>
是模型作者的提示词,我只是加了个lora和调整了参数,不过用自己的提示词复现同等质量的图也不难。
负提示词:(worst quality, low quality, illustration, 3d, 2d, painting, cartoons, sketch), tooth, open mouth
Step:100(30就够了我只是有强迫症)
Sampler:DPM++ 3M SDE和UniPC
高分修复和后期处理没什么必要
Refiner 切换时机:0.01
宽高:768×1280
CFG:7
模型作者图种子:2516340429
我的图种子:3083935321
顺带一提,封面是DPM++ 2M SDE Heun,看了半天分辨不出来它跟3M哪个更好一点
纸上得来终觉浅,要学会跟着视频一起敲,要动手实操,才能将自己的所学运用到实际当中去,这时候可以搞点实战案例来学习。