Fooocus Tutorial Part 3 - FaceSwap/Pyracanny/CPDS & More Explained

Summary

在这个系列的第三部分中，我深入探讨了图像输入标签页，并分享了一些我学到的技巧以及一些可能你不知道的生活质量提示。我解释了一些刚开始使用时让我困惑的设置，并详细分析了控制网格及其设置，以帮助你完全理解它们的作用以及哪些设置更适合什么目的。从历史记录日志到如何使用输入图像和变化标签页的高级技巧，我涵盖了如何利用这些工具改善和创造图像，包括如何调整表情和合并多个图像以产生创新的视觉效果。此外，我还介绍了不同控制网的特点和它们如何影响最终生成的图像。

Takeaways

• 📖 这是关于Focus系列的第三部分，深入探讨输入图像标签页的技巧和生活质量提示。
• 🔍 历史记录日志功能允许用户查看过去的生成记录和使用的设置，方便重复使用。
• 📅 历史记录默认显示当天的图像生成，但通过Focus的输出文件夹可以查看更早的生成。
• 🔄「加载参数」功能让用户可以快速应用以前的设置到新的生成中，但不包括输入图像的设置。
• 🌈「非常细微」的设置用于在图像中添加轻微变化，如改变表情。
• 🖼️ 图像提示功能允许将多达四张图片合并，以创建新的图像。
• 🎨 控制网的设置（如Pyan和CPDS）有助于在生成图像时保留原图的某些特征或风格。
• 🔄 面部互换功能专注于检测并应用图像的面部和头发到最终生成中。
• ⚖️ 「权重」和「停止于」设置影响控制网的影响力和应用时长，从而细致调整最终图像的生成。
• 👁️ 结合不同的控制网和文本提示可以创造出丰富多样的图像效果。

Q & A

什么是输入图像标签?

输入图像标签允许用户上传自己的图像,AI 会使用这些图像的样式、人物、构图、颜色等要素来影响最终的生成图像。

Pian控制网的作用是什么?

Pian控制网会提取图像的线条图,保留图像的细节,然后应用到用户提供的文本提示词上,从而在生成新图像时保留原图像的某些结构。

面部置换是如何工作的?

面部置换会检测图像中的面部和头发区域,然后将其应用到最终的生成图像上,无论是只使用文本提示词还是结合图像、控制网和文本提示词。

权重和停止步数设置如何影响控制网?

权重决定每个控制网对最终图像的影响力度,停止步数决定每个控制网应用于多少个生成步骤。合理设置这两个参数可以平衡不同控制网的作用。

生成图像的步数对图像质量有什么影响?

步数越多,图像的质量可能会更高,但需要的生成时间也会更长。默认性能设置下的步数为30。理想情况下,找到步数与质量及时间之间的平衡点。

Outlines

00:00

😊输入图像选项卡的提示和技巧

本段概述了一些使用输入图像选项卡的提示和技巧,包括历史日志的使用以及如何复制和查看以前的生成设置,微调表情的方法,组合使用多个输入图像的控制网等。

05:02

😃控制网设置的详细解释

本段详细解释了输入图像选项卡中的各种控制网设置,包括图像提示、Pyan、CPDS、面部交换等,以及权重、停止等参数如何影响控制网的作用程度。

Keywords

💡历史记录

历史记录是一个功能，用户可以通过它找到之前生成的所有作品。这在视频中被提到，用于回顾和重新利用旧的设置和输入。例如，如果用户创建了一个他们喜欢的图像并想要再次使用相同的设置，他们可以通过历史记录来复制这些设置到剪贴板，然后粘贴到输入框中，这样就可以重现之前的生成条件。

💡输入图像

输入图像是视频中讨论的一个关键特性，允许用户上传自己的图片作为生成过程的一部分。通过使用输入图像，用户可以指定一张或多张图片，以此来影响最终生成的图像。例如，用户可以上传一张人物的照片，并要求AI改变该人物的表情，或者将其放置在不同的背景中。

💡控制网

控制网是视频中解释的一个高级功能，它包括一系列的设置，如Canny、CPDS等，这些设置帮助用户精确控制图像生成的过程。例如，Canny控制网可以捕捉图像的细节并将其应用于文本提示，而CPDS则使用一种保持对比度的去色方法来影响图像。

💡变体

变体是在视频中提到的一个设置，允许用户在生成图像时添加微妙的变化，如改变人物的表情。通过调整变体设置，用户可以轻松地生成具有不同情绪或特征的相似图像，例如，从一个微笑的表情转变为大笑。

💡权重

权重是一个控制参数，决定了每个控制网对最终图像的影响程度。在视频中，通过调整权重，用户可以精细控制如图像风格、颜色等元素的强度。例如，增加一个控制网的权重会使其特征在最终图像中更为突出。

💡停止点

停止点是一个设置，它定义了控制网在图像生成过程中应用的阶段。通过设置停止点，用户可以控制某个控制网的应用时间，例如，设置为0.5意味着控制网将在生成过程的前50%中起作用。这允许用户在保留一些原始特征的同时，引入新的元素。

💡面部交换

面部交换是一个特定的功能，它检测图像中的面部和头发，并将它们应用到最终生成的图像中。在视频中，这个功能用于将一张图片中的面部特征转移到另一张图片上，同时保留或改变背景。这对于创造具有特定人物面部特征的定制图像非常有用。

💡图像提示

图像提示是视频中解释的一种方式，允许用户上传多达四张图像，并将它们的特征结合到最终生成的图像中。这是通过分析上传的图像的风格、色彩、构图等来实现的。例如，用户可以结合使用几张图像的风格来创造一个全新的图像。

💡Canny

Canny控制网是一种特定的算法，用于捕获图像的线条和细节，并将其应用于生成过程中。在视频中，通过使用Canny，用户可以确保生成的图像保留了原始图像的某些结构特征，这对于需要保持图像细节一致性的场景非常有用。

💡CPDS

CPDS（对比保持去色化结构）是一个由Focus团队创建的自定义控制网，它结合了深度控制和去色化方法，将图像转换为黑白，并使用该结构作为控制网。这种方法与Canny不同，但同样用于精确控制图像生成的过程。在视频中，这被用来影响图像的风格和结构。

Highlights

历史日志允许您查看并复制先前的生成设置

非常微妙设置可用于略微改变图像表达式

图像提示结合文字提示可生成混合图像

平安 NET 可捕捉图像细节并应用于新图像

保真去色网格可提取图像结构作为控制网格

人脸替换 ONLY 替换人脸和头发细节

权重控制网格的影响程度,停止控制应用的步骤数

示例展示不同设定下的图像生成差异

高级设置允许组合多种控制网格

控制网格可保留原图元素同时变更风格

人脸替换可与其他控制网格及文字提示组合

停止设置控制网格的持续步骤百分比

权重和停止共同控制每步影响程度

默认设定通常效果最佳

组合使用控制网格及文字提示创造新图像

Transcripts

in this third installment of my series

on Focus I want to dive deeper into the

input image Tab and show you some tricks

I've learned as well as some other

quality of life tips that you might not

know about I will explain some settings

that even confused me when I first

started to use them and break down the

control Nets and their settings to help

you fully understand what they are doing

and which ones are better for what

purpose if you haven't seen my previous

videos feel free to start there if you

like otherwise let's get started first

thing I want to show you is something

some of you may not know about and that

is the history log click Advanced then

click history log in here you can find

all your previous generations you can

use this if you created something in the

past and wanted to know what settings

and inputs you used if you want to reuse

one of these settings all you have to do

is click copy to clipboard then hit okay

go back to the Focus window and paste

that into the text prompt you will

notice the generate button changed to

load parameters this will change all the

settings and focus to the ones you used

for that previous image one thing to

note is that while it loads all the

settings if one of these images was

generated while using the input image

settings it will not load in the

previous images you were using you will

have to read those manually the second

tip about the history log is that you

may have noticed that it only shows the

images you created that day you can see

all the images you have created if you

go to your focuses output

folder go to your focus outputs folder

then you can see where your Generations

are separated into folders by date when

you go into a folder you will see a log

HTML file double click that and it will

open that days log file where you can

see those

Generations if you don't see a copy to

clipboard in a past log file it was

probably made before that option was

implemented this is something added

recently okay I'm going to reload my

page to go back to default settings and

we will now go into input image in the

upscale or variation tab I mainly want

to talk about the very subtle setting

this is a nice tool to add a little

variation in your image one thing I use

it for is changing the expression of an

image this can also be done in the

inpaint tool but it can come out a

little more natural here sometimes I

will load an image in here and select

very subtle for this I'm using all the

default settings other than changing the

resolution to 768

x344 and unchecking random with this

image we will try to get a happy

expression sometimes a few simple words

will

work and that definitely worked we got a

little smile but if we want to go bigger

maybe try to get her

laughing and still a subtle expression

so we can try going bigger adding mouth

wide open lots of teeth showing and see

what that gives us a little better but

again this is subtle change so it's it's

not going to go

crazy you can obviously go the other way

and get an angry

expression it's different with each

image some pictures don't take much to

get exaggerated emotions While others

can take some real

effort the rest of these settings I've

covered before so I'm going to move on

to image

prompt here we can combine up to four

Images together first thing to do is

click the checkbox advanced so we can

get these extra settings to show okay

I'm going to do my best to thoroughly

explain all these settings in here and

hopefully it will make a bit more sense

to you the first is image prompt this is

very similar to Mid journey in that when

you upload your own image and use that

as a base it will take everything from

the image the style character

composition colors and use it all to

influence the final image I'll grab an

image and add a simple text

prompt and we will see what we

get using this prompt you can see it

mixed a little bit of everything from

this image with our prompt now how much

it influences has to do with the stop at

and wait settings but we will go over

those a little

later next we have

pyan this is a great setting if you are

trying to get a character with a similar

pose or keep details from one image to

another what it does is is creates a

line art picture that captures intricate

details of your image then applies that

to your text prompt and any other

control Nets you

have so let's try this out I will add

this text prompt a woman in a business

suit in an

office

using this text prompt I got a

composition very similar to the image

even the background shapes are included

and then altered enough to match the

scene so with canny you can keep an

image and character similar to what you

have and still change the entire style

and come out with some great results

it's a very useful

tool cpds stands for contrast preserving

decolorization structure sounds complex

but it is a custom control net by the

focus team that incorporates depth

control combined with a decolorization

method that converts the image to black

and white and uses the structure of that

as a control net to influence an image

pyan and cpds are doing the same thing

but using completely different methods

you can successfully combine all of

these control Nets but when it comes to

these two it's best to only use one or

the

other face swap will detect only the

face and hair from an image and apply it

to the final generation either by using

just a text prompt or a combination of

images control Nets and text

prompts let's add another image

here and we will be adding this face to

this image structure for this we will

use pyate canny and on this image we

will select face swap without any text

prompt the AI will take the structure of

this image add this face and then make

up its mind to fill in the

blanks it will not add this face onto

this exact image as the control net is

only going by the overall structure but

with a text prompt describing a subject

like our first picture then we get a

composition very close to this

image but really you can make this go

wherever you

want with this text prompt a clown woman

at a

circus you can see we have the structure

of the first image the face of the

second and now our setting is a circus

you can see it's trying to put clown

makeup on her but the face swap settings

are pretty high since we want to keep it

close to this face but we can keep

adding to the text prompt to get more

changes to the face or even start to

lower the face swap settings a little

eventually pushing the text to start

overtaking the control net we can also

um add an image prompt

here let's say we like the color style

of this

image and want it to start mixing in a

bit since we already have two control

Nets I will increase the weight for this

to help it start showing without a text

prompt once again the AI will do its own

thing but you can see it's adding those

bright colors

in now that you better understand what

each of these control Nets do let's talk

about how these settings work and affect

them here are the settings we can change

these will directly affect whichever

control net we have

selected weight is how much we want the

selected control net to influence the

final image the higher the weight the

more it will push that control Net's

influence per step the stop at takes

that control net and the weight and then

applies it to the gener a for however

many steps we set it for with

Performance Set to speed all images will

go through 30 steps that is how many

times the image will run through its

algorithm until the final image is

completed you see that every time you

generate

images think of the stop at setting as a

percentage with the pyan here at 0.5 it

is 50% so with 30 steps total this

control net will be applied for the

first 15 Steps only basically it will

stop at 50% this gives the image a

chance to get its basic structure formed

and then at 50% let the other control

Nets and text prompts continue to

influence the image in whatever way you

are asking them to setting it higher

will have the control net applied to

more steps and keep the structure

likeness towards the original image and

will give the AI less ability to draw

Outside the Lines the same goes for

setting it lower that will give more

freedom to the AI and not adhere to the

structure so much so the weight is how

strong the control net affects every

step and the stop at is how many steps

the control net will run for there is no

magic number between the settings but

going off the defaults that each control

net starts with is best we can delve

deeper but I feel like any further and

it will be overwhelming but I hope this

is starting to make sense before

wrapping this up I want to show some

visual side by-side examples of

different settings with stop at and

wait I didn't run cpds in this vid but

as you can see it's similar to

pyan and here are a few featuring face

swap and I think I will conclude this

video here I don't ever ask for likes

and subscribes I figure you will do that

on your own but I will say thanks to

those who do I appreciate it and it

motivates me to keep creating I will see

you in the next one