All language subtitles for 3. Working with Loras - SDXL

Send this video we're going to do a couple of Transformations we're going to transform

This image here

Into this image

Wish we've already done so that's not much of an achievement

But we're also going to

Create something that looks a little bit like this

Out of that image so the starting point is this workflow this is a workflow that emerged in the early

How many days of STX l

And is one of the workflows that

I think was really designed by someone who

Automatic 11:11 cuz it's got

Interesting features in it that automatically 11:11

What we want to do is to use a Laura file to create this type of image and

It's going to be the paper cut one

So we'll see how to set things up we'll do a transformation on the workflow and then a transformation

Turn off the images

Now the problem with the workflows that I messed it up a little bit it's all kind of disorganized

We can go here and we can choose the light theme

And we can actually see what things look like

Box here inside of coffee U

I want to move everything into that box

But I want to change the flow of things so we've got the refiner

Which comes second right at the top look at this image is beautiful we've got this

Hourglass we've got the city inside The Hourglass

Another city and a clock inside our glasses it's amazing what stable diffusion comes up with

And this is done using a very annual to dogs workflow

But what we want to do is to move everything down here but we want to

Put the bass right at the top

And we want to put the refiner at the bottom

And I also want to

Just organize things a little bit better

Now with a comfy your eye

Generally a good idea to work from top left to bottom right

You don't have to things will work just fine if you're working from base from from the lower

Left to the upper right

But that is in general you want to be able to move things around easily so let me show you how to do that first

That's a very quickly here we got the bass

Model hair is 0.9 as we need to change that to 1.0

Go to sampler

We've got the initial image

It's a little bit messy I'll give you

The starting point and also the ending point for this particular edit so you've got to workflows that you can you

We've got this refining output here is grouped

And we've got the refiner checkpoint here that's that's in a group as well

So just go ahead and move this

Stuff around we want the refiner to be at the bottom

So I'm just going to move this one here

Awesome

And then we'll move this guy up here

Maybe even resize this

So that is a little bit larger

1080 / 1080 so it's actually

Quite a lot larger than you can see them on the on the screen

We just

Speed through rendering this is

Pretty quick and it's a lot quicker than automatic 11:11

Let's keep the bass up here for the time being but let's move the

Checkpoint down here

And then let's move the

Define sampler down here as well

Now we've got an unusual Arrangement which

Is this

Play texting codes

Links and will shift select those to

And we shifts like we should be able to move

Both of them at the same time that's it

Awesome

And now

Let's move

We've got this group here which is sdx 0.9 which is the

Title for the entire workflow I think

Let's change that to prompts and let's use this as a group

Play prompt

And we can now just resize this

And we can bring the

Positive prompt up here

We need a little bit of space the negative

Generally the advice is to keep the negative prompts fairly simple in

Is the Excel which is what we're going to do here

Just a couple of words that will do

And when the group together we can move them together like that

Things are looking a little bit neater a little bit more organized

You got a Laura here which

Don't really want

Because that's kind of like

Spoiling the game

You want at the Laura a little bit later

Now I know that there are some vaes down here so we can actually move these out of the way

Maybe move the vaes out

And I just collapse them by clicking there

We won't be having much to do with these ones in this

Inside of the

Come for you

And that's looking good

That's looking good

Let's change the colors back to dark

Now this is going to be one of the workflows

Well this is going to be like the midpoint in the workflows

If I do a renter at this stage let's see if it actually works if I've made a mistake okay

We need an input here

And is telling me that we don't have an input and I think that's because we got rid of the Laura file so let's just connect

Positive

Prompt

And will connect this as well

Now we're working with a 1080 by 1080 image

And the reason is 1080 by 1080 rather than 1024 by 1024 is that I found

Simple change in the size producer radical change in the image that we were getting

And I prefer the 1080 by 1080

Now with automatic 11:11

We really shouldn't be going below 1024 by 10:24 once you get down to say about 5:12

With sdx cell we should not

Down below about

1024 by 1024 if we go down to about 5. 1 2 5

1.5

Things begin to look weird and

You should try to stick around that one megabyte range

You can't work with logic file sometimes but

1 megabytes seems to be a sweet spot

Now this particular workflow is a bit unusual in that is is using a standard case sampler for the

And it's also using a standard case sampler for the refiner it's not using an advanced

What's casein

The author of this decided to use a low noise low noise level for the

For the refiner

Little bit of a waste of the

Role of the refiner stop reading the Orthodox way

So we're going to change the method so that uses an orthodox method

Initial one it's 20 steps it looks fine it works just fine I'm not going to be changing that

What will do is to

First of all maybe even now at the Laura file

But before we do that this project explain what's happening with these

Eclipse texting codes

Because we've only got two prompts we've got one negative and one positive

They're both being used for the

Checkpoint

Here in the checkpoint there how do we do that

How do we arrange that

This is something you might see from time to time

And you might want to replicate it so I'll show you exactly how we do that

Now what we have here is a collapsed unit collapsed node

If I am in right click let me double click and come here which is clip

And we choose text to conditioning

That gives us one of these boxes

And these boxes need to be attached to a primitive so this is actually a primitive

And the Primitive is cold up by going to add node

Utilities

Primitive

And primitive doesn't have any kind of role to play

What we need to do

Is to give it a role to play so if we connected to the text

That connection will convert it to one of these boxes

Now for whatever reason

Whenever I try to do this

Come for you I start throwing a hissy fit I think it's maybe to do with the software that I'm using to record

Or maybe they may be an error inside the coffee UI but if I try to connect this now it will just

Play Jason Arrow so I'm not going to try to do that

But essentially to produce one of these units

That's the that's the workflow

You would then connect this to

Connect to this input here to this

Connect the text here

That would convert this to this guy here and I think what they've done here is just basically

Go to colors and change the color

And

Think

Went ahead and just

Play some rounded Corners to to the edges there so some of the edges

Little bit rounded I prefer to do that so

That was my contribution and some of the edges of square obviously

But I think just basically

Drag connect

Play that should turn it into a

Into one of these bad boys hit into one of these prompts

One of these texts boxes

Obviously this guy here would then connect to the clip

Coming from the model

And this then would connect to

Either the positive or the negative

Depending on which prompt you connecting it to

I'm a

Play Can't demonstrated more than that because like I say my software begins to

Throw a glitch every time I do that

And it allows you to have just two prompts

That we use both for the for the

Refiner and the base model

Now the next thing I want to do is to add a Laura file because that allows us to see

How Laura's work inside of sdx

Involves using a work at trigger word and we're using the

Trigger word which is papercut

Change the look and appearance of the results from

Something that was like

To something that's more like this

Did that

Let's add a Laura file so the Laura file

Is attached once again between the

Checkpoint at the sampler

We just drag out

The note from the model

We get Laura Lauder

And we choose paper

Now papercuts is in the design folder that I created earlier

Will my system

And then we connect

Connect to the model here and that would disconnect this link here

And replace it there we are

Then we need to connect the clips otherwise we're going to get an error and because we're working with

This model and the sampler

We need to work on these two Eclipse that are connected to the sampler so let's open them up

On collapse

Mike's

And then I think the approach is going to be

To take this clip to the Laura

Clip

Like so

And then take this clip

Bring it down here

And then take this guy bring it down here

And that should complete the loop for the

And also for the models of the model going up here

The eclipse going up here

The model coming down here and the clips coming down here

This Behavior where we've got the text being brought in from

A proceeding

That's something I don't think we've seen too much of that but that's something to bear in mind that sometimes

Comfy you are it allows you this kind of backward bleeding at work flow where you bring things in

And this kind of way

Sometimes that's really useful for organization

And in actual fact for this one it just makes things a little bit neater

Just having two sets of prompts

Okay now the other thing I want to do is to change the behavior of the refiner

We want to actually just see what's going on a little bit more so it will change back to spline

And then change it back before it drives me nuts

We've got a BAS

Make a sampler which is running

Are you able to see the settings when you look at the file but what I want to do is to change this from a standard

Sam + 288 in advance sampler

So we're just going to go ahead and delete that

And then we'll double click

Case sampler advanced

That's the one we want

And then we can begin to connect everything up now so we need to connect up the

Blatant

Connect the latent to this latent here

And then we need to connect the eclipse so we can bring that up bring this down

Open it up

Open it up

And then this guy here is coming from here

And let's bring that down a little bit

And it's collecting data from the positive so we connect that to the positive

Which must be in this guy has negative

And that's looking good we can collapse those two again

And then where is the model is all kind of a link together

Down here with the refiner sampler the advanced sampler

Connect those two up and then we need to do the older dogs

Arrangement in terms of the samplers

Now the Orthodox Arrangement is basically about

Steps to about one step down here

We're going to go with something a little bit different just a little bit of an alteration to that

Schedule will go with the

What's on the

And increase the steps there to 25

And that means we do the first 20 steps and then we do the last five steps

That ratio of

421 that's going to be good for this image I think

And I'm just going to move the

Seeds up a little bit

Maybe fix it

Like that

Let's run this and see what happens

Now I think what will happen is that we won't get an image here that's because the

Open that up

Connect the latent to the samples there

Once we've done that it should run so what I'll do is to cue the prompt

And maybe we'll change things back to the streets mode

Will you be able to see here is that with

Comfy you are it runs fairly quickly

I even went dealing with the 10:24 or 1080 by 1080 image

It does run very quickly

Once you get to very large image sizes you begin to get

What's this sort of accelerating reduction in speed

Auntie is exponential

Exponentially

Increase thing as you increase the the size of the images

For a standard 1024 by 1024 even with my screen recorder

Using the the graphics card

It runs pretty quickly and this is one of the big advantages

List of using

Fusing

A comfy UI

Over automatic

Make 11:11

So you can see that we were able to take a workflow that was essentially an automatic 1111 work

I'm very very quickly change it so that it follows the more Orthodox method

Of having a sampler and an advanced sampler

Inside of coffee UI

Having a beginner phase for the base model and an ending phase for the for the refiner model

And that is I would say this looks good

Zoom in that looks like a nice

Type of paper

Design

And if we compare it to some of the earlier ones that we were looking at before

I like this one here

With all the swirly stuff happening this one is a beautiful one but with this one we didn't have too much guidance did we

But we could always come down here

The designer of the of the sdx

325 megabytes

But the way come for you I handled it and just ran through it very quickly was was kind of amazing

What we could do is to come to

I hear we could

Click on the

Settings there and see maybe some of the settings that people have used

And you can see if we open this guy up here

That's beautiful

Papercut Angel made of paper okay maybe that's a prompt

What kind of prompt we need to use in future

And you can see that we didn't really need to add this angle bracket stuff in I don't think it's actually doing very

Very much

You could just go ahead if we wanted to

And change the settings with the strength of the clip model

Now if you're wondering why we've got the model and the

It's kind of accessing is is influencing both the model the checkpoint

And also

It's accessing the clip

Play the clip Network the

The neural network

You can change the settings for both of them and it will influence the outcome that you get

I think with this one a setting of 1.0 work very well

And one of the things I noticed

With this one was that the clock face we actually had one two three four

456

This is unusual with stxr it doesn't always get the numbers right

And with this one maybe because clock face is a fairly standard it was able to interpret the clock face correctly

That's impressive

And I really like the outcome

So that basically is how we would transform a

Set alarm an orthodox automatic 1111 image the image workflow to a

STX L workflow inside of comfy UI

When I say standard is basically the workflow that you'd find

Being advocated by the design by the developers who initially came up with the idea

Of having multiple models

And we also looked at how to reorganize the space to make it a little bit neater

And a little bit more logical

And hopefully when you get these files you'll be able to play around with them and to see the changes that I made

I need and how everything works and links together

So look out for the files inside of the resources

And I will

And by saying I think it's a great idea

To go to

Civic a

Look for a few more Laura's that would really work well with your kind of workflow

And try to

It's not getting stuck in using them and seeing how well they can work

Are the ones that I like this one here they can really transform

Your images and they can produce a really nice result with very little work

Inside of comfy you

So I think this is a great resource to make you

In your creative Journey