Checkpoint、Lora、Hypernetworks是 Stable Diffusion Web UI 生图的重要工具,它们有各自的特点,结合不同的生图场景选择一个或者多个叠加使用,能够更好的命令 Stable Diffusion 生成理想状态的图片。
以人像生图用通俗的方式解释
checkpoint :必不可少的模型,全局生成模型,图像的生成全依赖它。
Lora:可选的模型,局部生成模型,用来控制某些部位的生成,如:人脸、手指、服饰等。
Hypernetworks:可选的模型,用来给整个图片进行风格重塑,如:雕塑风、摄影风等。
通过生成一张人像图,来简单看一下 Checkpoint、Lora、Hypernetworks 在生图中的作用。
Checkpoint
原图参数
1girl,sweater,white background,
Negative prompt: (worst quality:2),(low quality:2),(normal quality:2),lowres,watermark,multil figures
Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 12, Seed: 3882543293, Size: 512x512, Model hash: 7c819b6d13, Model: majicMIX realistic 麦橘写实_v7, Denoising strength: 0.7, Hires upscale: 2, Hires upscaler: Latent, Version: v1.9.4
LORA
Lora - 人像 lora
追加 Lora ,其他参数都不变,提示词增加: <lora:Mio_1.0:1>。
参数如下:
1girl,sweater,white background, <lora:Mio_1.0:1>
Negative prompt: (worst quality:2),(low quality:2),(normal quality:2),lowres,watermark,multil figures
Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 12, Seed: 3882543293, Size: 512x512, Model hash: 7c819b6d13, Model: majicMIX realistic 麦橘写实_v7, Denoising strength: 0.7, Hires upscale: 2, Hires upscaler: Latent, Version: v1.9.4
Hypernetworks
Hyperentworks 真人摄影风格。
追加 hypernetworks,提示词增加: <hypernet:hypernetworkForProject_hypernetworkForProject:1>
参数如下:
标签:Diffusion,Web,生图,quality,Hypernetworks,Checkpoint,Model,Lora From: https://blog.csdn.net/qq_23202687/article/details/1435799791girl,sweater,white background, <lora:Mio_1.0:1>, <hypernet:hypernetworkForProject_hypernetworkForProject:1>
Negative prompt: (worst quality:2),(low quality:2),(normal quality:2),lowres,watermark,multil figures
Steps: 20, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 12, Seed: 3882543293, Size: 1024x1024, Model hash: 7c819b6d13, Model: majicMIX realistic 麦橘写实_v7, Denoising strength: 0.7, Lora hashes: "Mio_1.0: 438992b461c0", Version: v1.9.4