Original subtitles

With many artificial intelligence

applications, we are all now producing

images. At least we are trying. I also

use many of them for both my own work

and experiments. Each has its own

strengths and weaknesses. In this video,

I will talk about an artificial

intelligence application that produces

very successful, very realistic images

that can address several important

issues at once. The main issue is the

successful AI image production tools are

paid. They're right because producing

anything with artificial intelligence

requires significant computing power.

There's no point in me agreeing with

them. After all, these are all

commercial ventures and naturally they

want to make money. The easiest way to

do these operations without spending

money is to place our own computer at

the point we call that computer. In

other words, to leave all the load on

our own computer. This leads us to open

source software. We don't need to pay

anything for these. Completely free.

They are there for development and

continue to be developed. After reaching

a certain point, companies that wish can

switch to a paid version. In fact,

OpenAI has such a story. When they first

started, as the name suggests, OpenAI

began with this open software. Then they

evolved into a company extremely secure

open software. I have been using them

for a long time. I haven't seen any

problems. Since it's open, if there were

any virus, no one would install it or

those who know would immediately see it

and request its removal. These types of

open software are generally hosted on

GitHub. In the past, installing these

open software on our computers required

some technical knowledge. Now with the

Pinocchio software, that's no longer

necessary. You can handle everything

very easily. By the way, the biggest

advantage of generating visuals on your

own computer is that you don't face any

copyright issues. The secondary solver

of this software is character

continuity. It's one of the most

frequently asked questions I receive.

So, how do we ensure the continuity of a

character we've created? We created a

female character. It also solves how to

place this character in different

settings like in Ulude during winter

when it's snowing on the beach in summer

at the cinema or in a studio. By the

way, I couldn't help but wonder if this

topic is related to a news article I

read recently. Someone created an AI

influencer named Itana and it's earning

$11,000 a month. It's a figure that

boggles the mind. I also found the

Instagram account of the AI influencer

and it's still active. In summary, with

this technique, you can create such a

character and maintain its presence in

various environments. By the way, I

think 2024 will be the year of AI

influencers. I can almost hear you

saying, "Who cares about your opinion?

Just send the goods." I'm starting right

away. I'm on the

Pinocchio. As always, I will add all the

links and prompts I use in the video

description. I'm clicking the download

button. On the page that opens, you can

download Pinocchio according to the

platform you are using. There are

platform options available such as

Windows, Mac, M1, M2, M3, Intel Mac, and

Linux. Since I am using a computer with

an M1 Apple silicon processor, I click

on the M1, M2, M3 Mac link. On the page

that opens, I click on the link click to

download Pinocchio for M1, M2, M3 Max.

It started downloading the latest

version. The download is complete. I

double click on the file I downloaded to

open it. In the window that opens, I

first drag the Pinocchio application and

drop it onto the applications folder.

Then I right click on the patch command

file and click on open link from the

menu that appears. A warning appears

saying the developer is not verified. I

know it's not a problem. I press the

open button again. The terminal

application opens on my computer. Here

it asks me to enter my computer's

password. I will enter it, but I want to

clarify one point here. In the terminal,

the cursor does not move as you type. It

looks like you're not typing, but you

actually are. I enter my password and

press the enter key on the keyboard. As

you can see, the process completed.

Notification appeared. Now I return to

the application folder and double click

on the Pinocchio application to open it.

Pinocchio has opened. From this page, I

can choose where the Pinocchio

applications files will be stored on my

computer and select a light or dark

theme. It can stay like this. I press

the save button. Now my Pinocchio

application is ready. From here I click

on the visit discover page button or the

discover icon over there and all the

applications I can use within Pinocchio

have opened. There are many applications

like invokeai, stream diffusion, dream

talk. I explained in detail what we can

do with many of them in a previous

video. I'm leaving the link. You can

watch it from there. These applications

don't have a direct connection with

Pinocchio. You can think of Pinocchio as

a player. In short, it lets you use

whichever application is available on

the counter. All the applications within

Pinocchio are on GitHub if you want

instead of using Pinocchio. You can go

to GitHub, download the files, download

the necessary software and versions to

run the application, run the software,

and set up all the settings to use the

artificial intelligence applications. Or

instead of all this, you can set up the

Pinocchio I showed you and have it do

these processes for you. I'm returning

to the discover page. The AI app is

Focus. Yes, it's spelled with three O's.

I clicked on it. On the page that opens,

there's a link to the GitHub project.

Again, as I mentioned earlier, those who

want can also set it up from here.

There's also information that Focus uses

the stable diffusion infrastructure.

There's also a link to the Twitter

account of Cocktail Peanut, the creator

of Pinocchio, which you can follow if

you want. I'm clicking on the download

button. As you can see on the page that

opens, many software components like

Git, Zip, Node, JavaScript are missing

for Fukus to work. Thanks to Pinocchio,

I don't have to deal with where these

are located, where to download them from

or which version to

download. I click on the install button

and start downloading all of them. The

downloads are complete. I click on the

okay button. Now we can move on to

Fukus. Just now when I click to install

Fukus, Pinocchio helped me download the

necessary software to use Fukus. Now I

will download Fukus. I click on the

download button. It downloaded. I click

on it again. From the page that opens, I

click on the install link. From the menu

on the left, a warning came up under the

installation required heading that

another software is missing. I click on

the install button again. The

installation of the missing brew

software has also started completed. I

click on the okay button. Now the source

files for focus have started

downloading. These are quite large

files. Please wait patiently for them to

download. I must reiterate that you are

downloading the entire AI model to your

computer. I mentioned this before.

Download the model to your computer,

open it in Pinocchio, and start running

it. If you're sure it's working, you can

turn off the internet and continue with

visual production. So, in summary, you

download it, but you only download it

once. After this, you can create

unlimited, free, royalty-free

productions as much as you want. The

download is complete. I click the start

button from the menu on the left. From

the open submen, I can start using the

mode or rather the model that I want.

Today, I will proceed with the realistic

mode. I click on the realistic mode. I

started downloading its models. The

model is about 6 and 1/2 GB in size. I'm

waiting for it to be downloaded as well.

The download is complete. To be sure, I

restarted Pinocchio again. I click the

start button from the menu on the left

and select the realistic mode from the

opened submen. I saw the message app

started successfully. The processes are

complete. If I want, I can click on the

link here to open the focus application

in an external browser or I can click

the web UI button from the menu on the

left to have the focus application run

directly within

Pinocchio. The most important feature of

focus as I mentioned at the beginning is

close-up or focus which I guess is where

the name comes from. It produces very

very realistic visuals. Right now,

without touching any settings and using

the default settings it opens with, when

I paste a prompt into the box and

produce any visual, you will see that it

creates incredibly realistic visuals.

I'm pasting my prompt right away. A very

beautiful Turkish girl fitness at the

gym. I meant to say a beautiful Turkish

girl doing fitness at the gym. Let's see

what kind of visual will come out. By

the way, you absolutely need to enter

the prompts in English. It doesn't

understand Turkish. You can use deep l

for quick translation. By the way, I

don't want the same sexes. Since the

topic started with itana AI influencer,

I went with women. Of course, my female

viewers can also produce visuals of

handsome men. As you can appreciate, it

would look a bit odd for me to produce

visuals of handsome men. Did I seem

homophobic this time? I wonder. It's

very difficult, you know. Anyway, I'm

continuing. We will see how

knowledgeable our AI model is about

Turkish girls. I'm pressing the generate

button and waiting. This waiting is

related to the graphics card power of my

computer or rather your computer.

Computers with powerful graphics cards

are faster while those with weaker

graphics cards are slower. You can also

see it step by step. I speed it up in

editing. Of course, I don't have such a

computer. The first image is completed.

I'm clicking on it and enlarging it. How

is it? Not too bad. I guess it looks

quite realistic. The hands turned out a

bit problematic. I think the background

and face came out nicely. The left arm

looks a bit injured. I guess it happened

while lifting weights. Whatever

happened, the second one forms, changes

clothes, it drew a model of a typical

Turkish girl. The second one is also

completed. This one looks nice, too.

We'd have a hard time distinguishing

whether it's a photo or AI generated. We

know there's no such person on Earth.

Let's enter another prompt. This time,

let's not make it a person, but a room

or room in a house. I want to show you

different things. A 1960s style room

with orange objects. I mean, I'm saying

a 1960s style room with orange objects.

Let's see what will come out. I'm

pressing the generate button and

waiting. It seems a nice room is coming.

Let's see it complete. The first one is

completed. I'm clicking on it to open. I

think it turned out nice. It has the

feel of 1960s homes we see in movies.

It's predominantly orange, just as I

wanted. If you want to save it, you can

click on the download icon here to save

it. I named it 01 and saved it in PNG

format. The second one doesn't seem bad

either. It looks like it won't have a

television in it right now. There will

be a fireplace. It's done. This is

another house interior. So, they

considered the reflection of the mirror.

I'm saving this as 02png. Now, let's

start examining the features of focus.

For this, I'm clicking on the input

image checkbox. Details have opened up

at the bottom. I have four options here.

We will discuss all of them in order.

First, let's look at the upscale or

variation tab. By uploading an image to

the drop image here section, we can

create slightly varied versions with

various strongly varied versions with

very strong and enlarge them with

upscaler. I should mention that the

upscaler here does not enlarge the image

by stretching it. It recreates the image

using the uploaded one as a reference.

Let's do an example. I click on the drop

image here section. Let me enter the

1960s orange room image I just created

for my computer. What is its size? for

example, 1.3 MGB. Let's go check its

dimensions in Finder. I right click on

the image and select get info. It's 896x

1,152 pixels. I go back to Pinocchio and

upload my image by clicking on drop

image here. Then I click on upscale to

2x from the options below. I press the

generate button and wait. Meanwhile,

every time you try something different

on focus, it will download other models.

Wait patiently for it to download. Since

everyone uses it for different purposes,

it doesn't download the entire package

at once. That would be silly. The sizes

would become quite large. If you're not

going to upscale at all, why would you

download its model completed? It created

the first one. I click on it and open

it. It has become much higher quality

compared to the previous one. Let me

download it to my computer. I click on

the download icon and save it as zero-

qpng. I immediately check Finder. I

right click on the first image and

select get info. Its size was 1.3

megabytes with dimensions of 896 x 1152

pixels. I right click on the new one and

select get info. Its size has become 4 7

megabytes. The dimensions have also

changed to 1792. By 2304 it was this.

Now it's this. It's not just enlargement

but reinterpretation. It improved

sections. It corrected the distortion in

this corner as well. We will continue

from the input image section and show

the advanced section because the other

tabs of the input image are linked to

the advanced section. I click on the

advanced checkbox and the advanced

settings open the performance section.

Here I have three options. Speed,

quality, extreme speed. In other words,

when the AI creates photos, it goes step

by step like in stages. It draws

something more at each stage. The more

stages there are, the higher the

quality. Think of it like sanding. When

you sand it 30 times, it becomes shiny.

The speed option completes the visual in

30 steps. Extreme speed completes it in

eight steps. Quality completes it in 60

steps. You can proceed with whichever

one you prefer depending on the power of

your computer's graphics card. I would

like to clarify this. The quality

difference related to this visual is

actually a difference in AI usage. In

other words, 60 steps mean thinking 60

times. Eight steps mean thinking eight

times. As the steps increase, you give

the AI more chances to think, but

naturally the time also extends. Under

performance, there are aspect ratio

visual dimensions. It supports all

visual dimensions included in stable

diffusion. Since we're here, I'm

choosing 1024x 1024. The image number

determines the number of visuals it will

generate each time I press the generate

button. By default, it comes as two. If

you want, you can increase this number

up to 32. I'm leaving it at two. The

random checkbox allows the seat value to

be assigned randomly. If you uncheck it,

you can enter the seat value manually.

The second tab is styles. This is

actually the strongest aspect of focus.

There are over a 100 styles you can

select and use whichever you want. When

you hover over these styles, a pop-up

cat image appears. From these images,

you can roughly understand what that

style does. For example, focus cinematic

is a cinematic cat image. There are

bokeh lights and such in the background.

What else? Adorable 3D character,

meaning a cute three-dimensional

character. The watercolor style, those

like this. All of these are styles that

have undergone very extensive training.

For example, the watercolor style. I'm

not entirely sure, but I guess it

completed its training by scanning

hundreds of thousands of watercolor

images. Car advertisement images, for

instance. It has been trained with

thousands of car advertisement photos.

To use them, simply click on them to

activate. The other tab is the model

which is the section we selected when

opening focus. We are currently working

with a realistic image but you can

change it from here as well. In the

advanced tab there is a gagen scale. As

you increase its value the cleanliness,

vibrancy and artistry increase. It

becomes more beautiful yet there are

nonsensical hallucinations. I generally

try not to exceed a value of seven. But

make sure to try different values

yourself to gain experience. I can't

change it right now because we're in

extreme speed mode. I'm going back to

the settings. If I set it to speed,

meaning if I take it to 30 steps, it

will be activated. Image sharpness is

already understood from its name.

Increasing it enhances the sharpness.

Now, let's create an image using styles

from the styles MK. I'm choosing Fango.

I'm entering my prompt. An appetizing

fruit platter of summer fruits on a

wooden table. So, I want an appetizing

fruit platter of summer fruits on a

wooden table. Aspect ratio, meaning the

dimensions. I choose

1,344x74 pixels. I'm not changing the

guidance scale. I press the generate

button and it starts creating.

Completed. Let's see. Very stylish. Like

a Fanggo painting. If Fango saw how

easily and quickly this was done, I

think he would cut off another ear.

Normally, we enter prompts in writing,

but we can enter them with a visual

prompt in Focus. In fact, we can also

control what Fukus should do with these

visual prompts. I activate the input

image checkbox to be able to upload my

image. Then I go to the image prompt in

my second tab. I open the advanced

checkbox located at the bottom of this

tab so I can see all the features. Now I

click on the first box and upload one of

my own photos from my computer. I also

open the advanced settings. I choose the

image size as 1024x 1024 pixels. I

switch to the styles tab. I remove the

default ones. I have the idea of

robotizing my uploaded image. For this

reason, I type robot into the search

box. A futuristic cybernetic robot

appeared and I select it. There's a

futuristic biomedical cyberpunk option.

I select that too. Now I've reached the

place where I can adjust the impact of

the uploaded image on the new image that

will be created. First, there's the stop

it option. Stop at Look, this is I just

had an epiphany. Stop at Stop it. Stop

the engine. Stop at it comes from here.

Look, I just figured it out. Stop. I

mentioned that the image is created in

steps. For example, speed is 30 steps.

Stop. It determines where to stop

reading the image during these 30 steps,

thus affecting the newly created

image. For example, right now it's 0.5,

meaning half. If I set speed to one, it

won't look at this image. After the

first 15 steps, it will work

independently. Or if I set it to one, it

will continue to read the sample image

throughout the entire production. on the

right side is weight. This weight

determines how much the new image will

be influenced by the image I added. I'm

coming to stop at. I set it to 0.8

because I want the influence to continue

for a long time. I'm not entering a

written prompt and I press the generate

button. It's completed. It was

influenced by the blue t-shirt. Since

stop at is high, it also resembled me.

Since the weight is in the middle, it

was able to change the background and

such. We've reached the point I

mentioned at the beginning of the video.

Let's create a character. Let's create

visuals based on the clothes this

character wears, the place they go, and

the time. I access the advanced

settings. I ignore the input image.

First, I select the quality. I will

create my own character, so it should be

high quality from the start. The

existing styles are sufficient. I remove

the negative and added masterpiece. Also

reviewing the dimensions of the visual

that will be created. It's 1:1, so it

should be 1,024x 1,024 pixels.

I increase the guidance scale a bit from

the advanced tab. I want a more vibrant

visual, so I paste my prompt. Close up

portrait of a very beautiful brunette

girl with blue eyes taken. Video in

photo studio, meaning a close-up

portrait of a very beautiful brunette

girl with blue eyes taken in a photo

studio. Desertia. Artificial

intelligence continues. When producing

visuals, it can't fully convert indoor

lighting to outdoor lighting or vice

versa. For this reason, it's important

to decide on this when creating the

initial prompt. Will it be indoors or

outdoors? I'm pressing the generate

button. Let's see what kind of result it

will give. It's important for us to see

the face of the visual clearly. This

will be very important when creating new

visuals. The first visual is completed.

A beautiful girl was created. Looks

real. It's almost impossible to

distinguish this from a photograph. This

is good for me. I'm downloading it to my

computer. One of my predictions for 2024

is artificial intelligence that can

produce videos indistinguishable from

reality like the one we just saw in the

photo. Therefore, this technique will

become even more important. Then,

instead of just photos, we will produce

videos with our character in different

outfits and various locations. I'm

activating the input image and coming to

the image prompt section. I'm uploading

the model image I just downloaded. I'm

also opening the advanced option in the

image prompt section. There are four

options here. We've talked about the

image prompt. There's also Pyrokini,

CPDS, and phase swap. I'm choosing face

swap. We will discuss the remaining two

as well. If you notice, stop at

automatically became 0.9. By entering a

prompt, it wants the image I'm going to

create to take this image's face to the

final step. I'm boosting it even more.

I'm setting it to one. It doesn't stop

influencing and is present in all steps

until the end. I'm entering my prompt

fitness at the gym. Pink hair. So, it

will be a girl with pink hair working

out. I'm setting the performance to

speed. I click generate. Let's see how

it turns out. The first one is

completed. The image turned out nice.

The fingers are a bit problematic. It

would be better if she didn't wear

earrings at the gym. But we got the

image we added. I'm saving it to my

computer by clicking the download icon.

The second one is also completed. It

looks better since there are no hands

around. Let me try another image. I'm

entering my prompt. A pose in an elegant

red evening party dress, blonde hair. So

I said a pose in an elegant red party

dress, blonde hair. I'm pressing the

generate button. First one done. Very

stylish. There's a bokeh effect in the

background. The girl looks quite pretty.

The second one is also completed. This

time focus thought of a sitting pose. I

think this one turned out nice, too.

Let's try another example. I'm entering

my prompt right away. A girl in a beret

and coat walking down the street on a

very cold snowy winter day. So, a girl

wearing a beret and coat walking down

the street on a very cold and snowy

winter day. I'm saying generate see what

comes out the first one is completed. I

think this one turned out very nice,

too. Coat, beret, a snowy street there.

Second one done. Nice pose, too. Let's

diversify a bit more. For example, we

saw a photo and the character in that

photo struck a pose. We want our

character to strike the same pose. Let's

see what we'll do. I have a pose here.

The model has struck a pose like this. I

want my character to do the same. I'm

turning to focus. Let's keep our face

swap photo. I'm not changing it. This

time I'm choosing Pyraini. Pyroini is

copying the movements. I'm going to the

second photo box. I'm adding a stance

pose here. I'm uploading my image. It

will only take the character stance. I'm

not touching the settings. I'm leaving

it as default. I'm not entering a

prompt. I can enter one if I want. just

the visual dimensions. I'm choosing a

vertical image. I click the generate

button and it starts. It will take the

face from the first image and the pose

from the second image. Let's see what

comes out. The first image is completed.

Since I didn't enter a prompt, it did it

on its own. It could take the face and

pose as they are. There's a bit of

distortion in the arm. The second pose

is also completed. I think this one

turned out better, but there's still a

problem with the hand. Let me try

entering a prompt. I said it on the

beach in summer. I say generate. The

first one is completed. I think it

turned out much better when I entered a

prompt. The eyes are a bit problematic.

This became a good example. I will show

how to fix these. The second one is

completed. The face and eyes are nice.

It just looks like there's a bit of a

spinal issue. Anyway, we'll try it out.

We'll continue until we find the one we

like the most. We've reached the last

section within the image prompt. CPDS,

which stands for contrast preserving

decolorization structure. It's not

something very important. Unfortunately,

you are also exposed to my obsession

with trying to do everything perfectly.

I get this silly feeling that if I see

it there, I should explain it. I should

tell what it is. But there's nothing to

be done. As I said, you are also exposed

to this. In short, it extracts the

contrast from the image you add and

applies it to the newly created image. I

check the CPDS check box. From the third

photo edition box, I upload the same

pose photo from my computer. Now, it

will also take the contrast of this pose

photo. I select the dimensions of 768x

1,344 pixels from here. And again, I

press the generate button without

entering a prompt. Let's see what will

happen. Completed. As you can see, the

colors have become nicer because the

other one is a real photo. This one was

influenced by it. This is the other

visual. I think both turned out very

beautiful. Within focus, you can modify

visuals, expand them, and intervene in

selected areas.

I'm coming to the third tab within the

input image the unpaint or I'll paint

tab. Let's start with the first example.

There was a snowy street visual I

created. Let's take that for example.

Drop image here. I click on the click to

upload link and upload the visual from

my computer. I can immediately say to

extend this from the left and right

under the outpaint direction heading. I

check the left and right checkboxes.

Then I just press the generate button.

Let's see what happens. The first one is

completed. As you can see, it expanded

the visual beautifully. There's nothing

disturbing at all. As you might guess, I

can also expand by entering a written

prompt. So, prompts I write can be added

to the areas that will be expanded.

Like, I don't know, add a tree or add a

street lamp. Here, we can make small

aesthetic adjustments with focus.

Painless and effortless. I'm in the in

paint or out paint tab. I immediately

select improve detail from the method

section. We had created an image with

problematic eyes. If you remember, let's

make an adjustment to it. I click on the

click to upload link. I take our patient

with the problematic eyes from my

computer. I need to mark the problematic

area here. From here I can see the

shortcuts to use this canvas. I can also

adjust the size of my brush from the

right side. Now since the eyes are

problematic, I will select them by

painting over them. I hold down the

shift key and scroll the mouse wheel.

The image zooms in. I paint the eyes

with the precision of a doctor. All

right, I remove the zoom with the R key.

There are also quick prompts here. It

seems like the AI knows where it has

issues. I click on the beautiful eyes

link. It added it to the prompt section.

Nice. I scroll up and press the generate

button. Look, it zoomed in on the area I

drew in my workspace and I can see the

work there step by step or rather stage

by stage. As you can see, it corrects

the errors. The first one is completed.

The problems with the eyes have been

largely resolved. So, what else can we

do? We can add, remove, or transform

something. I'm returning to the in paint

or outpaint tab. I erase the eyes I drew

by pressing the silicone. I select

modify content from the method section.

I will try to change the bracelet on the

arm. I zoom in on the image again. I

select the bracelet here. In other

words, I paint the bracelet. I completed

it. I exit the zoom by pressing the R

key on the keyboard. I come to the end

paint additional prompt box and enter

the red wristband prompt. I press the

generate button. Let's see if the

bracelet will change. The preview of the

bracelet has appeared. It's creating it

slowly. Only that area is visible

because I selected it. It's completed.

As you can see, the red wristband has

appeared. It also matched with the

shorts. From here on, you can do

anything you want. For example, you can

change the hair color to pink. We've

reached the final section. In this

section, you can add an image or an

animation image and have it convert it

into a text prompt or in other words

into text for you. It reads the image

and converts what it understands into

text for you. I'm in the describe tab.

Here we can have it describe a photo or

anime image that I input. I clicked on

the click to upload link and edit my

model from my computer. Let's see how it

will describe it. I'm clicking on the

describe this image into prop button. It

will immediately describe the image in

its own way and enter the resulting

text. Enter the prompt box above. Yes,

it wrote in the generate box. A woman

with a big earring next to blue eyes.

So, a beautiful woman with a big earring

next to blue eyes. Let's add another

example. Let me add the model we used

for this pose. I'm clicking on the click

to upload link and uploading the image

of the model I used for the pose. I

clicked on the describe this image into

prompt button. It responded, "A woman

with glasses leans against her neck

wearing a floral shirt." So I said, "A

woman with glasses is leaning against

her neck wearing a floral shirt." I'm

pressing the generate button. Let's see

what it will create. It didn't get so

she's not leaning there. It's like she's

placing her hand. There's a feminine

pose. It couldn't figure that part out.

Anyway, it created it. It didn't fully

understand or couldn't create the

visual. I don't know exactly about that

part, but I think the visual turned out

very realistic and beautiful. If you've

made it this far in the video, I

congratulate you. It's been a bit of a

long video, but I think if I'm

explaining something, I should touch on

everything. I'm still dragging it out

and can't stop. If you have any

questions, you can write them as

comments under the video, leaving my

social media accounts here. You can

follow me from there and ask questions.

That's it for this week. See you in the

next video. Goodbye.