Portrait Magic: Retain Faces in MidJourney Rendering

Summary

本视频教程介绍如何使用Midjourney AI工具,根据照片重新生成人像,包括上传参考照片,调整生成的参数,融合不同风格的图像元素,迭代多次渲染以使结果更接近原图,从而创作出颇具 cyberpunk气息的数字艺术肖像。

The video is abnormal, and we are working hard to fix it.
Please replace the link and try again.

The video is abnormal, and we are working hard to fix it.
Please replace the link and try again.

Outlines

00:00

🎥 介绍利用Mid-Journey生成照片风格肖像的视频内容

本段介绍了视频的主要内容:将利用Mid-Journey这个工具,基于一张真人照片,生成不同风格的肖像。会展示不同的参数设置,如背景、发型和服装的变化,推敲生成肖像最佳的图像权重参数范围,使生成的肖像尽可能保留原照面部特征。主要目标是让AI生成的肖像尽可能贴近原始图片。

05:00

📷 准备原始照片素材

本段说明了准备原始照片素材的方法:在Discord中上传要生成肖像的照片,右键复制图片地址,在Mid-Journey中粘贴图片地址,设置宽高比例参数,这样可以为AI提供原始照片作为生成肖像的参考对象。为了展示图像权重的不同设置产生的效果,设置了5种不同的参数进行渲染,以选择最佳参数。

10:01

🔎 分析不同图像权重设置的效果

本段分析了5种不同图像权重设置生成的肖像效果。权重越低,生成的肖像越不像原图;当权重≥1时,生成的肖像面部特征越接近原图。但权重太高时,也会限制了创作的灵活性。需要在保真和创新之间找到平衡。图像权重不仅用于脸部特征,也可提供全身立绘参考。

15:02

🎨 合成图像背景

本段展示了如何通过Blend模式,合成自定义背景。直接混合2个图像时,不太能保留面部特征。但将肖像和背景分开处理,先生成肖像,再混合背景,可以获得较好效果。也可以在Prompt中描述元素位置,直接指导AI进行合成。

20:04

✏️ 调整Prompt提升识别准确率

本段介绍了使用Mid-Journey的Describe功能分析输入图像,获取AI对图片的描述,帮助完善Prompt提高生成质量。可以结合图像、描述、权重、自定义修改,逐步调整Prompt,使输出肖像不断贴近预期。

25:06

🚀 综合运用多种技术生成理想肖像

本段总结汇整了前面各段介绍的技术,包括上传原图、获取描述、设置权重、合成背景、调整Prompt等,层层渲染生成最终肖像。遵循提供的步骤,可以制作出风格独特且保留面部特征的肖像。

Keywords

💡人像

这是视频的核心主题,指的是在Midjourney AI中生成具有真实人物面部特征的虚拟人像画作。它反复出现在脚本中,说明了如何保持生成图像中人物面部的逼真程度。

💡背景

背景指的是人像图像外围的环境元素。视频讲解了如何为已生成的人像画作更换不同的背景,例如黑客风、蒸汽波等风格的背景。这体现了Midjourney的强大之处。

💡着色

着色就是为图像添加颜色。视频中演示了如何通过改变 prompt 指令,为已生成的人像画作更换不同的服装颜色,以丰富图像的风格。

💡图像权重

图像权重(image weight)是Midjourney中的一个参数,范围从0.1到2,用于控制生成图像与样例图片的相似程度。视频详细解释了不同权重对生成结果的影响。

💡描述

Midjourney中可以使用"描述"(describe)命令来分析样例图片的内容和风格,生成一段对图像的文本描述。视频展示了如何利用这些描述来优化人像生成的prompt。

💡融合

指将两张图像融合成一张新图像的技术。视频介绍了多种在Midjourney中实现图像融合的方法。这可以用来将生成的人像与新背景进行融合。

💡叠加

叠加指在Midjourney中反复基于当前结果生成新图像,逐步优化结果的技术。视频建议多次叠加不同图像,使生成的人像不断趋于逼真。

💡放大

放大是指增强图像分辨率的过程。视频中在获得满意的低分辨率结果后使用放大,以获得高清的最终人像图像。

💡prompt

Prompt是Midjourney中用来输入生成指令和样例图片的文本框。视频讲解了如何通过修改prompt的文字说明来控制生成结果。

💡逼真

逼真指生成结果与原始照片样例的相似程度。增加图像权重和叠加次数可以提高生成人像的逼真度。这是视频教程的核心目标。

Highlights

我们将使用 Discord 上传参考图片

调整图片权重可以控制生成图像与参考图片的相似程度

描述功能可以让 AI 分析图像并生成提示

混合模式可以将两个不同的图像组合在一起

将肖像和背景分开可以使混合模式效果更好

我们可以通过多次渲染和图像叠加来逐步完善肖像

上传参考图片、调整权重和生成描述可以制作更准确的肖像

可以修改生成图像的样式和细节

可以迭代多次渲染、调整细节和图像叠加来逐步改进

每次迭代可以越来越聚焦于保留原始肖像的面部特征

描述可以提供AI对图像的理解,帮助优化提示

逐步调整权重、提示和样式可以生成理想的肖像

多次图像叠加可以使生成的肖像更准确逼真

保留面部特征同时改变背景和样式可以创造全新肖像

通过不断调整和迭代可以生成高质量定制肖像

Transcripts

in this video we're going to work on a

portraits inside the mid-journey and

specifically take a photo reference keep

it all characteristic of face and

creating the portraits inside the mid

Journey with the different settings

example right here we heard some steam

cyberpunk kind of backdrop replacement

you'll notice they're all same change a

little bit playing on a hair changing

this color also change some little bit

outfits as well

we'll try to also push as far as we can

till face become less recognizable so we

see what ranges we need to stay with

and of course work more on the

backgrounds and change colors and

outfits as well as we're going cologne

and of course our main goal be sure they

are

person face purchase will resemble the

original image as closest as possible

start work on the project we need our

original photo that we're going to

reference that AI will be trained and

use it because me Johnny utilize Discord

so we're going to use as well right in

the bottom in the chat you'll notice you

have it plus symbol click on a plus and

select upload file

navigate to the file or portraits you

want to use it and select after

selection we show you press enter

that will show this image was uploaded

to you on your server best way to verify

this if you go and click on an image and

it will open in a preview and this is

actually what we needed because we need

open one image open in preview right

click on a Windows says copy image

address

on a Mac I think it just copy image

and this is how in Discord will go to

reference so let's go ahead and do next

we'll go

tap imagine and in a prompt we paste the

path to our image this is a Simplicity

things we can do and next we're going to

type you can type for example

um so next what I want to do is actually

preserve our ratio and because I don't

know at yet what is fully correct ratio

of this image I will just says okay

let's do portraits for this and we'll

set two by three this this simplest way

we can create image reference

um portraits reference to our image that

we uploaded

okay and here we have our first render

even we can render it does somewhat

resemble some elements but will last our

glasses we lost some

options and this is okay for the basic

but you notice it is not come closer to

what we wanted

so we have other options we can apply it

we can actually tell me Journey how

closer we want to resemble to the image

so what I'm going to do is copy this

line that we have it right here

and I'm going to paste this below

next what I want to do is tell me

Journey how much of the image I want to

effect how weight of the image

importance of the image so for this

we'll go to type space Dash Dash

IW and this is stands for image weight

and then next usually you type any Image

Weight and it can be from 0.1 up to 2.

but I will utilize another options

because I want to see how the all image

weights will affect I'm going open curly

brace and type 0.1 this is the lowest

image weight can be comma we'll type 0.5

comma 1. comma 1.5 comma 2. so what is

happening here I'll tell mid Journey

render same things for me but render

first time with image weight 0.1 next

time 0.5 next with image weight one it's

what we have right here by default it's

one next 1.5 and 2. so it just will give

me five different options and I can see

how much image weight is important to be

my portrait closer to what I wanted so

let's go ahead press enter

and you'll notice right here it should

come and ask if I want to

the process five right here says you

imagine you have five prompts on your

template let's go ahead and click yes so

it will start processing five jobs now

this time you'll notice it is a three

started because I am on a standard and

two more and then queue so we'll go

ahead and wait slightly um till it's all

the completed and we're going to review

them

okay render completed let's view our

result the first we're going with is 0.1

it's at the lowest and you can see it is

hardly resembled at all Only We Know It

is the female

with a portraits about hell body up to

waste and also take his rock cover red

dress so but it does not even come

closer to our original portraits if

we're going in step to 0.5 this little

bit better because we have some pauses a

little bit closer resemble but again

the porches face not even close to what

we have in a region when we're going

with weight one you'll notice now we

start having glasses because to

recognize some of the images

a little bit more closer to what we have

in the original but it's still a little

bit

away from what we wanted

and when we cross this one point weight

one now it's definitely image more

influence so right here we can see

definitely number one facial it's very

close and so is the number three

actually they're all somewhat closer

except number two a little bit more far

away but generally there are much closer

I will say almost is working for as a

portrait based on his face and when

we're using the maximum 2 you can see

this is very close to other portions to

the sample very well the problem is when

we're using

number two for example as the weights

weighted two then we start having more

heavy influence

of the

portraits for of the image for example

right here even when we specify red

dress with does have a red outfit but

it's also have a green outfit it does

try to compensate with putting like red

background but it is Image Weight it's

more heavier so you have less

flexibility and this is where you come

with fine tuning how well you will go

how much details you want to keep it and

how much details you want to lose but

again this is just our start point

because we're going to work on this and

add little bit more

so till we jump into other things let me

Point like for example right here we

have a 0.1 image

and in some cases it's hard to see why

do you want to use the image in this

case but

um let me give you one of the very good

example notice we have a health body

shot and female so it does reference

pauses and everything

example right here I have an image full

body shot and for example I want to

create the image with a full body of the

girl dancing or other things in this

case for my AI this will be reference

it's what I wanted I want full body shot

and I know some people was asking me how

you can create a full body shot because

it's a cut off maybe a feet or other

things so this is way to do it so let's

go ahead and I will say copy image

address for this

and we're going to imagine we'll paste

this image next let's go say Rococo

style so we'll just describe what we

wanted here and remember what we wanted

we want ER 2x3 just in this portraits

everyone Image Weight 0.1

okay so right here render we can look

and notice this is 0.1 we have it full

body almost shot also position of the

hands and this is their lowest effect of

the image and we can expand more we can

increase more weights for example right

here a little bit higher and you can see

we start having better example as well a

little bit the lick of the light and of

course the more we're going

the better it is

influence of the image so you don't

necessarily need to have it all have

influence if you don't really worry

about the facial expressions but this is

also will provide you additional options

to create for example let's say I'm

going here and I like some of the style

let's go with this one so I would like

this image and I'm going to upscale this

and now I want to use the actual blend

these images and we come to another step

of our techniques blending images

and blending images we can take two

different images and combine them

together let AI decide for this so this

was kind of relatively new at to my

journey but it does help us to create

portraits but if we want to work with

the porches here we need to pay

attention to kind of tricky situations

so for example let me show you and we're

going to use the blend for this mod

it will ask for several images you can

go up to I think six images

when it's selection we need to go drag

and drop or click and upload our images

and I'm just going to select two of

those images we have it

and let's press enter and now if we look

on our blending result

it does not resemble the original

portraits because it took two images

and combine them together we do have a

girl with wearing glasses like in three

but the facials somewhat resemble but

again the hair and other things is

definitely took from the other image

what I found out it's better on a

blending Mode work if you have adjust

the portrait and separate different

background so here is example let's do

two different images again we'll use the

blend mode by the way you can have it

way more if you just select and you can

add more images but we'll just use it

too

and we'll have it one as a portrait and

another one separate as the backdrop

notice on our backdrop we don't have it

any person so AI does not have any

reference to the what person it look so

it's definitely will take only from

first one and now we can just go ahead

and press

enter so it start processing and after

process is finished you can see right

here we have it same portraits the

person is same as a portrait and just

apply to this background that we have so

it's actually does a very good job

it my limitations with this

um descriptions because I needed provide

what background I wanted and after

margin which is not bad because you can

actually create background inside the

mid Journey specifically what you wanted

and reference later this in your um

blend mode with a portrait you wanted

again the purchase that you will blend

notice is still wear green code so we or

green sweater so we need to change this

and another nice options inside the mid

Journey that we don't necessarily need

to use it blend mode we can also take

two images and blend and kind of merging

them together

okay so here's the example I have one

portraits and I'm going to copy image

address on this we'll go type imagine

we'll paste the first image we'll go

open another image

copy this image address

space this image address

next we want to put description in a

blending again it's work very good it's

very interesting but here we can also

put a description where do I want it and

I'm just going and copy this description

including a spec ratio

so we can go ahead and paste in

okay I think right now we're ready to go

ahead and render

okay so right here our result and you

notice we do put it together but we have

a problem when it's image does not

resemble

we can overcome this much easy if we

just also apply some weights to this so

for example right here with image weight

2 and you can see we have quite a bit

closer now

we have our background and we have our

image with the model

so this way we actually can work but you

notice the

prompt it's have the big portion playing

in what we're creating

inside the mid Journeys have done one

excellent options to help us see what is

on our image to work on our prompt and

this is Will describe so let's go ahead

say describe and now notice we have it

one image that we need to upload to let

me Journey analyze this and I'm going to

use the image of our model

and we'll go press enter not as a

description it's given us information

the reason is why I'm like to use a

description of one it's provide for us

information it's how the mid Journey

seeing the specific portrait so if you

want to have something closer it will

respond

as well we have our respect ratio here

so let's have just example I'm copying

just only the prompt first prompt and by

the way you have four different prompts

and a notice going from more accurate to

more kind of flexible but you can try

either

and if we just type this prompt and give

it to us and process let's see how

closer it may get to our description of

the image we have it notice we don't put

any image reference inside it's just

pearly the description

and right here we have it from our

description so now we can go and

actually utilize both of them

description and text so let me go ahead

right there we'll click on our image

copy image address go down below let's

go click imagine

will paste our text next I'm going to

copy the prompt we have it

and we'll paste The Prompt as well here

so this time we should have our image in

the very close to the prompt we could

actually just even say Image Weight 1.5

to be sure we have it as close as

possible between image uploading

description prompt and also wait on an

image okay so right here we have our

image and you can see the facializer is

sampled very close I mean the pauses

everything

this is in a way we can now modify so

what I like to do we'll go copy of this

prompt including our the image reference

we'll go to type imagine we can paste

here and inside the string I can modify

for example wearing glass grass and

let's switch

cyberpunk

turtleneck in a style of the festive or

let's go I have again cyberpunk we'll

just

cyberpunk as much as we can okay you'll

notice we have it all this information

don't like hip hop style let's go put a

cyberpunk

I think

it should get message I already three

times put it cyberpunk because hopefully

it is there so now we can modify

and we'll see what has come up again 1.5

that should work fine

and right here we have our image it's

much different style I like it but look

at this it's facial very close on our

portrait

and I remember we did before we done

some images overlay ended with this like

cyberpunk we can to this as well here

for example if I like this I can go

upscale my image that I like it and we

can merge that with another image as

well to apply even extra steps and

so in this case we can go ahead

let's go type imagine okay and we're

going to take

our image that we used before this one

we'll copy this image again we'll take

our source

let's go paste our source here and now

we'll take this image they're creating

we can copy this image address

a little paste here

and we'll take now

this line that we have

copy and we can paste so what we've done

with this we say utilize these two

images but even more importance put it

on an image so we get our portraits even

closer but we still have it keeping all

changes that we wanted

so let's just in case per woman wearing

glasses

cyberpunk outfit

and style of the

rain so I'm changing slightly we can do

this one and maybe you know what let's

go just put it for fun

red outfit we'll see what it's come up

with

okay and right here we have our

combination notice how it's replaced

background the portraits all right or

that this one for example look very good

very close except we have broke a little

bit on our hands this one also look very

nice so you notice how we start adding

more and more layering to create the

portrait

so now we'll look on these different

techniques let's take them together and

put it and create the portraits of the

model in different environments we

wanted so right here first remember

we'll start with uploading our image and

I uploaded two different I also uploaded

background what we're going to use so

right here is our original image Next

Step what I want as preparation step

actually I want to go describe

and as described we're going to select

same image so we receive the

descriptions of the image that we

currently have

after this we're going to start

combining first we'll go type imagine

and we'll take a portrait of the

movement that we want copy so copy

address we go paste this address inside

next we're going to take their

descriptions copy and paste and

sometimes description change for example

right here somehow they figure out it's

a solve it kind of very fun for me but I

want to change some of the stuff for

example a young woman wearing glasses in

elegant sweaters so we'll leave this but

maybe in elegant sweater we switch to

elegant red

raincoat so we'll just switch a little

bit here

in a style of emerald bronze this is

actually our background we can switch

this to the coloring they wanted but I

think let's leave it right here maybe

you know in of cyberpunk because I'm

going to

that's what I'm going to use cross

processing meditation it's neon lighting

we leave this fist of armster

and it's that Christmas punk I don't

know what Christmas Punk we have it

cyberpunk okay and instead Soviet let's

go to Retro

so when we're done with this also

we'll set our Image Weight and again for

the image weight I'm not sure which one

I'm going so I'm going 1.2 1.5 and 2 so

we're going to see on three different

ways and usually I do this just to be

sure I have it right a selection because

sometimes it's very hard to say and it's

easy doing one pass to see what you have

okay let's go ahead press enter on this

and yes I do want to have three

different versions okay so right here is

render let's preview this is our number

two and you can see we have the green

let's go down and find right ones this

is a bit better rain or I like the rain

code this is actually looking very close

and most important our portraits is

getting quite a bit closer in this case

next it's kind of important you need to

select one that you like and resemble so

I'm going with this red one number two

we go upscale that

and we're going to actually merge a

couple more times so I'm going to take

copy of this cyberpunk kind of a

background so I'm going to put that one

here and next I'm just take from this

and copy all the way up to here space

and paste it so this should help us to

kind of create more interesting type of

the backdrop

actually what I did right here but

accidentally I grabbed the original

image it's not what I wanted so let's go

ahead again copy this image right here

we copy that one

okay so we go to paste

okay with the background I want to

actually copy this so I need to go copy

this image address

okay we'll go paste here and then I want

to actually copy this string because

otherwise I just copy original image

that I don't need it

okay right here and let's render well

just in case as the this one process we

also will see what's happening if we

just have the original image and okay so

right here our image it's completed this

is one that did it with the first it's

actually kind of interesting except the

face does not necessarily look very good

and this is another the problem when we

start merging with another background we

start losing a little bit more of the

originality uh well original faces and

what's happening we can always emphasize

a little bit more for example if I like

this background an image and everything

we can kind of blend more in more with

original photo

so again example we go take again this

first one we copy image that we already

have

and now we'll have it like another

process so it's almost you developing

any overlay over and over with the

different images as you need it

so again we'll copy this one now I'm

going to take and copy this image

remember it's have it background but

it's also have a little bit closer with

what I wanted

and we're going to copy

this part and paste

in here so let's go ahead

this will be our another pass so we have

it let's look what we have so far we

have our original pass with portrait

after we select portraits we have it

pass with the portraits and background

and now because it's already render we

have a pass of the merge

background with render portrait with our

original

and here we have a render and if you

look this is very nice this is one

resemble very good our original photos

some little bit different and it's

possibly a lighting but if you go

position of the head nose and eyes it

does work actually very very good so let

me upscale those two ones

and here's our render I'll say this is

actually resembled this is the resemble

very good and notice we change

everything we just keep it on your face

and put it in different environment

different lighting thank you for

watching this video