Stable Diffusion涂鸦生建筑效果图教程

一、演示内容：

1、SD的photopea插件涂鸦发送之controlnet，

2、用scribble+shuffle预处理，

3、zaha的lora，

4、提示词描述

所有的AI设计工具，模型和插件，都已经整理好了，文末扫码即可免费获取噢~

二、成果展示

白天

参数立面感

三、教程准备

1、一张草稿

2、一张shuffle参考图

本图为陈总自制lora模型，生成的售楼处的图

3、用到的模型，仍然是henmixReal_v40.safetensors，C站可以下到

四、生图步骤

1、填写如下提示词

parameters

Steps: 12, Sampler: DPM++ SDE Karras, CFG scale: 9, Seed: 2971866181, Size: 1024x512, Model hash: d1aaf70725, Model: henmixReal_v40, NGMS: 3, ControlNet 0: “preprocessor: invert (from white bg & black line), model: control_v11p_sd15_scribble [d4ba51ff], weight: 0.6, starting/ending: (0, 1), resize mode: Crop and Resize, pixel perfect: False, control mode: Balanced, preprocessor params: (512, 64, 64)”, ControlNet 1: “preprocessor: shuffle, model: control_v11e_sd15_shuffle [526bfdae], weight: 0.4, starting/ending: (0, 1), resize mode: Crop and Resize, pixel perfect: True, control mode: Balanced, preprocessor params: (512, 64, 64)”, CFG Rescale φ: 0, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: zahahadid(448e3b918283), AddNet Weight A 1: 0.35, AddNet Weight B 1: 0.35

2、生图参数如下

3、测试推荐种子：

2060827079

2060827087

3816377789

五、重点讲解：

1、photopea插件：

插件下载几安装地址：

https://github.com/yankooliveira/sd-webui-photopea-embed

食用方法，与ps基本相同，免费是全功能，有广告，可以编辑后可保存或者直接发送到生图界面

2、提示词提取插件：

1、wd14-tagger和clip-interrogator

https://github.com/pharmapsychotic/clip-interrogator-ext

https://github.com/toriato/stable-diffusion-webui-wd14-tagger

3、lora参数为0.35，不宜过大

4、shuffle参数为0.4

六、过程图分享：

henmixreal大模型生成的效果

动漫大模型：revAnimated_v122

parameters

a rendering of a building with large windows on it’s side and a city in the background,people walking around it on the sidewalk,a digital rendering,David Chipperfield,hypermodernism,vray render,architecture,bridge,building,city,cityscape,cloud,cloudy_sky,day,east_asian_architecture,house,mountain,no_humans,outdoors,road,ruins,scenery,sky,skyscraper,tower,tree,
Negative prompt: overexpousure, sketch,illustrious,logo,text, paintings, sketches, (worst quality:2), (low quality:2), (normal quality:2), lowres, ((monochrome)), ((grayscale)), watermark, signature, text, [( NG_DeepNegative_V1_75T :0.9) :0.2], high contrast,
Steps: 12, Sampler: DPM++ SDE Karras, CFG scale: 9, Seed: 2161132774, Size: 1024x512, Model hash: 4199bcdd14, Model: revAnimated_v122, Clip skip: 2, NGMS: 3, ControlNet 0: “preprocessor: invert (from white bg & black line), model: control_v11p_sd15_scribble [d4ba51ff], weight: 0.6, starting/ending: (0, 1), resize mode: Crop and Resize, pixel perfect: False, control mode: Balanced, preprocessor params: (512, 64, 64)”, ControlNet 1: “preprocessor: shuffle, model: control_v11e_sd15_shuffle [526bfdae], weight: 0.4, starting/ending: (0, 1), resize mode: Crop and Resize, pixel perfect: True, control mode: Balanced, preprocessor params: (512, 64, 64)”, CFG Rescale φ: 0, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: zahahadid(448e3b918283), AddNet Weight A 1: 0.35, AddNet Weight B 1: 0.35

未加shuffle的图

预处理器选none的情况

写在最后

AIGC技术的未来发展前景广阔，随着人工智能技术的不断发展，AIGC技术也将不断提高。未来，AIGC技术将在游戏和计算领域得到更广泛的应用，使游戏和计算系统具有更高效、更智能、更灵活的特性。同时，AIGC技术也将与人工智能技术紧密结合，在更多的领域得到广泛应用，对程序员来说影响至关重要。未来，AIGC技术将继续得到提高，同时也将与人工智能技术紧密结合，在更多的领域得到广泛应用。

感兴趣的小伙伴，赠送全套AIGC学习资料和安装工具，包含AI绘画、AI人工智能等前沿科技教程，模型插件，具体看下方。