A1111 - Best Advice for amazing Stable Diffusion Images

Summary

This video tutorial offers valuable tips for achieving stunning AI-generated images. It guides viewers on selecting the best models, utilizing negative embeddings, and leveraging various models and lora adjustments for optimal results. The host also explains how to use the interface effectively, including adjusting sliders for clip skip and choosing the right upscale models for high-resolution fixes. Additionally, the video covers the use of extensions like 'a detailer' for face tracking and 'control net' for body tracking, providing a comprehensive guide to enhance image generation.

Takeaways

• 😀 Choose the right AI model for better results based on community ratings, hearts, and downloads.
• 🔍 Read model descriptions and prompts to understand how to effectively use them for desired outcomes.
• 📚 Utilize negative prompts and embeddings to exclude unwanted elements from AI-generated images.
• 🎨 Explore different models like Realistic Vision, ref-animated, and deliberate for various artistic styles.
• 🔧 Understand the importance of settings such as sampler, CFG scale, and high-res fix for image quality.
• 📸 Use the 'after detailer' extension for automatic face tracking and higher-resolution in-painting.
• 🌟 Discover and apply settings from successful images by checking prompts and parameters used.
• 🔗 Recognize that models, lora styles, and embeddings must be compatible with the base model version you are using.
• 🛠️ Adjust the weight of lora styles to prevent overemphasis and achieve a balanced result.
• 🖼️ Use the image-to-image tab for making subtle changes to existing images, like altering ethnicity while keeping the same composition.
• 🔧 Employ extensions like 'a detailer' for face tracking, 'control net' for pose tracking, and 'Roop' for face replacement to enhance image detail and customization.

Q & A

What is the main purpose of the video?

The main purpose of the video is to share tips and tricks for achieving amazing results with AI, covering both basics for beginners and advanced techniques for experienced users.

Why is choosing the right AI model important for the outcome of the AI-generated images?

Choosing the right AI model is important because a better model generally leads to better results in terms of image quality and accuracy. The model's ratings, hearts, and downloads can indicate its popularity and effectiveness within the community.

What are 'negative embeddings' and how are they used in AI image generation?

Negative embeddings are small models trained on content that the user does not want to appear in the image. They are used in the negative prompt to ensure that the AI-generated image avoids certain elements.

How can one find suggestions for using a specific AI model?

Suggestions for using a specific AI model can be found by scrolling down on the model's page, where there is information on the best positive and negative prompts, as well as other settings like the sampler and CFG scale.

What is the role of the 'after detailer extension' mentioned in the script?

The 'after detailer extension' is used for automatic face tracking and in-painting the face in higher resolution, enhancing the detail in the facial features of the AI-generated image.

Can you explain the significance of the 'CFG scale' in AI image generation?

The CFG scale determines how closely the AI sticks to the provided prompt. A lower value allows the AI to take more creative freedom, while a higher value makes the AI adhere more strictly to the model's training data.

What is the 'high-res fix' and how does it improve the quality of AI-generated images?

The 'high-res fix' is a feature that allows users to upscale the resolution of an AI-generated image for better quality. It involves using specific upscale models and adjusting settings like denoise strength to enhance image details.

What are 'loras' and how do they influence the style of an AI-generated image?

Loras are smaller versions of models that can influence the style of an AI-generated image. They add specific details or characteristics to the image, such as additional sharpness or artistic flair.

How can one adjust the settings for the 'image to image' tab in the AI software?

To adjust the settings for the 'image to image' tab, one can change the prompt, adjust the denoise strength, and use various extensions like 'a detailer' or 'control net' to modify specific parts of the image.

What are some tips for using the 'in painting' tab to modify parts of an AI-generated image?

In the 'in painting' tab, one can use a brush tool to paint over areas they wish to modify. It's important to adjust the render size to match the area being modified and to set an appropriate denoise strength to allow for changes while maintaining image quality.

How can one ensure that the AI-generated image matches the desired ethnicity or facial features?

To match the desired ethnicity or facial features, one can use the 'image to image' tab to adjust the prompt accordingly and set an appropriate denoise strength that allows the AI to make the necessary changes while staying close to the original image.

What is the purpose of the 'Roop' extension in AI image generation?

The 'Roop' extension allows users to replace the face in an AI-generated image with a photo of a real person. This can be useful for creating images that have a specific person's likeness but may require the original image to be close to the body and head shape of the person for a natural result.

Why is it recommended to use a lower batch size or batch count for users with slower computers or older GPUs?

Using a lower batch size or batch count helps prevent overloading the computer or GPU, as it allows images to be rendered one at a time or in smaller groups, improving performance on less powerful systems.

Outlines

00:00

🤖 Choosing the Right AI Model for Amazing Results

The video script begins by emphasizing the importance of selecting the right AI model to achieve the best results. It suggests looking at community ratings, hearts, and downloads to identify popular and effective models. The script introduces the concept of models and luras (smaller model versions influencing style), and provides tips on using specific models like 'Realistic Vision' and 'ref animated'. It also discusses the use of negative embeddings to exclude unwanted elements from AI-generated images and mentions the significance of choosing the right sampler and CFG scale for high-quality outputs.

05:00

🔍 Navigating the Interface and Enhancing Image Quality

This paragraph delves into the user interface of the AI software, focusing on adjusting settings like clip skip and selecting appropriate vae models. It explains how to add quick settings for clip stop and sdvae, which are crucial for customization. The script also covers the high-res fix feature, which improves image resolution and quality using upscale models like '4X Ultra Sharp'. The importance of setting the right denoise strength for image rendering is highlighted, and the process of placing models, luras, embeddings, and vae files into the correct folders within the software is outlined.

10:01

🎨 Working with Models, Luras, and Embeddings

The script discusses the compatibility of models, luras, and embeddings with different versions of AI software, such as SD 1.5 and sdxl. It explains the process of adding negative embeddings to the prompt field and adjusting the weight of luras to achieve better results. The paragraph also covers the use of the CFG scale to control how closely the AI adheres to the given prompt, and introduces the image to image tab for making subtle changes to existing images, such as altering ethnicity while maintaining the same composition.

15:01

🖌️ Advanced Image Editing with Inpainting and Extensions

This section introduces advanced image editing techniques available in the AI software, such as the inpainting tab for making detailed changes to specific parts of an image. It explains how to use the 'only mask' feature and adjust render size for局部 image re-rendering. The script also highlights various extensions that enhance the AI's capabilities, including 'a detailer' for face tracking and replacement, 'control net' for body tracking, and 'Roop' for face replacement using photos. Additionally, it provides tips for users with slower GPUs or older computers on how to use tiling for upscaling images without overwhelming their hardware.

20:01

🛠️ Optimizing AI Image Generation for Different Hardware

The final paragraph offers advice for optimizing the AI image generation process based on the user's hardware capabilities. It suggests using the 'batch count' and 'batch size' settings to manage the rendering of multiple images, which can be beneficial for those with slower or older GPUs. The script also mentions using the 'SD upscale selected' script for tiling-based upscaling to减轻GPU负担. The video concludes with a call to action for viewers to leave a like and comment if they have further suggestions or enjoyed the content.

Keywords

💡AI Models

AI Models refer to the various artificial intelligence algorithms designed for specific tasks, such as image generation. In the context of the video, AI models are crucial as they determine the quality and style of the generated images. The script mentions checking ratings, hearts, and downloads to identify the most beloved models by the community, indicating their effectiveness and popularity.

💡Prompts

Prompts are the textual descriptions or commands given to AI models to guide the generation of content. The video emphasizes the importance of positive and negative prompts, which help the AI understand what to include or avoid in the image. For instance, the script describes how to use 'best positive prompt' and 'negative prompt' to guide the AI in creating desired images without undesired elements.

💡Negative Embedding

Negative Embedding is a technique used in AI where a model is trained to avoid certain features or elements in the generated content. The script explains that this is done by textual inversion, which is a form of training on what not to include in the image. The video mentions downloading negative embeddings and using them in prompts to refine the AI's output.

💡Sampler

A Sampler in AI refers to the method used to select elements from a probability distribution, which is crucial in determining the final output of the AI model. The script provides examples of different samplers like Euler a or DPM++ SDE, which are used to influence the randomness and detail in the AI-generated images.

💡CFG Scale

CFG Scale stands for Control Flow Guidance Scale and it is a parameter that influences how closely the AI adheres to the provided prompts. The video script mentions that a CFG scale of 3.527 is good, indicating that it is a recommended setting for achieving a balance between adherence to the prompt and creative freedom in the AI's output.

💡High-Res Fix

High-Res Fix is a term used in the video to describe a process of improving the resolution and quality of AI-generated images. The script explains using upscale models like '4X Ultra Sharp' for enhancing image details, especially faces, which are otherwise not rendered well at lower resolutions.

💡Loras

Loras, short for 'low-resolution artifacts,' are smaller versions of AI models that can influence the style and details of the generated images. The video distinguishes between different types of Loras, such as 'XL More Artful Laura XL,' which is designed to work with specific base models like 'sdxl 1.0' and adds an artistic touch to the images.

💡Extensions

Extensions in the context of AI image generation are additional tools or functionalities that can be used to enhance or modify the AI's output. The script mentions several extensions like 'after detailer extension' for automatic face tracking and in-painting, and 'control net extension' for body tracking, which allow for more detailed and controlled image manipulation.

💡In-Painting

In-Painting is a process where missing or selected parts of an image are filled in or modified by the AI. The video script describes using the in-painting tab to make changes to specific parts of an image, such as adding sunglasses or changing facial expressions, providing a way to alter existing images without affecting the rest of the scene.

💡Batch Processing

Batch Processing refers to the ability to render multiple images or variations at once. The script discusses using 'batch count' and 'batch size' to manage the rendering of images, which can be particularly useful for users with slower or older GPUs, as it allows for rendering images sequentially rather than in parallel.

Highlights

The importance of choosing the right AI model for better results based on community ratings, hearts, and downloads.

Understanding the use of models with detailed instructions and suggestions for positive and negative prompts.

The role of negative embeddings in refining AI-generated images to exclude undesired elements.

The significance of samplers and CFG scale in achieving high-quality AI image generation.

How to utilize the high-risk fix with specific models for enhanced image details.

Insights on using extensions like After Detailer for automatic face tracking and inpainting.

Recommendations for models like 'ref animated' and their compatibility with different VAEs and negative embeddings.

The application of different models such as 'Deliberate' for photorealism and 'Epic Realism' for natural lighting.

The concept of 'loras' as smaller model versions that influence the style of AI creations.

Guidance on matching the correct version of models, loras, and embeddings for optimal AI performance.

How to navigate the Automatic1111 interface for effective use of additional parameters like clip skip and VAE model selection.

The process of implementing high-res fix for image upscaling using recommended models like '4X Ultra sharp'.

Instructions on where to place different AI components such as models, loras, embeddings, and upscalers in the Automatic1111 folder structure.

The use of 'textual inversion' for negative prompts and how to integrate it with AI models.

Adjusting the weight of 'loras' to prevent overemphasis and achieve a balanced AI-generated image.

The function of the 'image to image' tab for making subtle changes to existing AI images.

How to use the 'in painting' tab for modifying specific parts of an AI-generated image.

The integration of extensions like 'a detailer' for face tracking and 'control net' for body tracking in AI image generation.

The use of the 'Roop' extension for replacing faces in AI images with real photos.

Strategies for handling AI image rendering on older computers or with slower GPUs using tiling and batch processing.

Transcripts

today I want to share with you some of

the best tricks to get amazing results

with AI this video is great for

beginners but there's also a lot of

stuff in here for the pros so let's get

started to get amazing results you want

to make the chop for you as easy as

possible and this starts with the choice

of the model because the better the

model is the better your results will

look of course there's a lot of

different models here so when you go to

a Civic AI you can also see here how

many ratings does it have how many

hearts how many downloads does it have

so that you can see which are the most

beloved models by the community those

are often working the best of course

there's different categories I want to

show you some of my favorite models for

example realistic vision is one that I

really like and here already I want to

show you something that is really

important to understand how to work with

such a model so for that you want to

scroll scroll down and here you have a

lot of information of how to use the

model for example here is a suggestion

for the best positive prompt all of

these different words that you want to

use next to the description of what you

want to have as the image then there is

also here a suggestion for the negative

prompt in here on what you want to use

for all these words so that you get the

best out of that model there's two

different suggestions in here so that is

pretty awesome and you can also see here

in Orange there is a negative embedding

this is a very small Model A textual

inversion as it is called this is

trained basically on the stuff you don't

want to have in the image and that is

then put into the negative prompt so

here you can click on that link and this

will then go to the page where you can

download the negative embedding we will

talk about it in a minute where to put

that then when you scroll deeper here

you can also see you have here for

example the sampler you want to use so

either Euler a or DPM plus plus sde

carers below that it says CFG scale of

3.527 is good and then also for the high

risk fix use the 4X Ultra sharp model

for that again I'm gonna show you in a

second where to put that there's a lot

of information here that you can use but

there is even more information that you

can use so when you scroll up you can

see these images and you can check out

the images that you like for example

this is really nice with the neon light

in the background so click on that and

when you click on that image you can see

here on the right side The Prompt that

has been used the negative prompt that

has been used the sampler the model the

CFG scale the steps that have been used

and the seed you can see here it says in

the first comment the after detailer

extension was used to create this image

that extension is something that does

for example automatic face tracking and

then automatically in painting the face

in a higher resolution so that's also

very nice I have a video about that

right here to check out that extension

other models I like are for example ref

animated now again you want to scroll

down and here on that page you'll find

something really useful so when we

scroll down again you find here a lot of

interesting information but on top of

that you see that this model works well

with the vaes orange mix and the klf8

anime 2 model and then also the Blast 2

vae so that you can use that below that

you find here the negative embeddings

that I am so fond of because they are

really helpful here's a very nice list

of them you can click on these links to

download them again I'm going to show

you where to put all of that in a second

so you really want to check out all of

that information and for each of these

models you want to scroll up click on

images you like here to look what kind

of settings prompt negative prompt has

been used to create that and you want to

start with this information and then

change the prompt to work your way into

that model to get amazing results right

from the start now there's other models

out there like for example deliberate

and this can be trained on different

styles for example this can do

photorealism and digital painting at the

same time there is also epic realism

which is really good has a little bit

more natural looking light so that's

also very nice and then another thing

that is really important is different

lauras you want to use now what loras

are is that they are smaller versions of

models and they can basically influence

the style that a model can create one

that I like really much I have a video

about that here is the XL more Artful

Laura XL also means that this Laura is

supposed to be used with the sdxl model

you can see here on the right side in

the information the type is Laura the

base model is sdxl 1.0 so that means

this does not work without SD 1.5 model

and it does not work with a St 2.x model

it has to be a sdxl model on the other

hand here we have another Laura it's

called a detailed tweaker now what this

does is that it adds some more details

to the image so that can be very useful

but this one as you can see on the right

side is for the base model SD 1.5 so not

for 2.x and not for sdxl and you will

find that most models and most lures at

the moment on cvdi are trained for SD

1.5 they have a lower resolution but

often still have a very good image

quality so you can absolutely start with

SD 1.5 and get an amazing results out of

that often even better than sdxl at the

moment because like I said there are so

many different loras for that so many

different models that are already

trained for that now let's have a look

at the interface of automatic 1111 what

are important things you might want to

know here so first of all we have up

here a slider for clip Skip and we have

a Chooser for the vae model you can see

I have for example orange mix and I have

the vae 840 000 and also 560 000 so I

can choose them here so they fit to the

model and I can play around with that

where do you get these sliders well you

want to go to settings and then on the

left side you want to scroll down until

you find user interface you click on

that you scroll down a little bit more

until you find quick settings and in

here you can add these additional

parameters so when you click in an empty

space you get a long list from where you

can can select these different editions

and what you want to add here is clip

stop at last layer and also

sdvae so these two choices when you add

them they will add these two additional

choices up here and then you want to

click of course on apply settings so

that the settings are saved and then

when you reload the UI you have these

two choices up here another thing this

is really important for good quality is

the high rest fix because usually with

the SD 1.5 model you are for example

rendering with 512 by 768 so with the

high res fix if you click on that this

is opening you have some additional

settings here you want to use some good

upscale models so I suggest to you the

4X Ultra sharp model the 4X nmkd model

and also the 8X nmkd model both of them

are super scary male versions I will

link all of that below the video and

then of course you want to upscale two

times and you want to set the denoise

strength between 0.2 or 0.35 not too

high because then it changes the image

too much although you can also

experiment with that now I want to show

you the difference this is an image I've

rendered with just the 512 by 768

resolution and as you can see the face

doesn't look too good the other details

are kind of okay but overall the image

could be better now here I have the high

res fix and this as I think we can all

agree looks amazing the face is fixed

everything looks beautiful this is not

using restore face this is just using

the high-res fix in that case I'm using

the 4X nmkd model so at this point we

have talked a lot about different models

and loras and embeddings where do you

put them in the automatic take 11 11

folder now the easiest one for that is

the negative embeddings these go into

the embeddings folder in here so you

just want to double click and then you

save them in here you can see I have

already a bunch in here then for the

models you want to go into the models

folder in here

and in there there's two important

folders there is this stable diffusion

folder this is where all of your models

go for normal rendering and then up here

you have the Laura folder this is where

all of your loras go so that's very

important other folders that are

important in here for you is the es

argon folder this is where all of the

different app scalar models go and then

down at the lower end you find that the

vae folder this is of course where all

of your vae models go there's another

important folder for you this is for the

control net extension because that has

its own models for control net so for

that you want to go into the extensions

folder and then in there you want to

look for the SD web UI control net

folder and in there you want to look for

the models folder and this is where all

of your control net models go I'm going

to provide a link under the video where

you find a lot of different appscaler

models here for example you can see the

the 4X Ultra sharp upscaler when you

scroll down a little bit more you see

here the 4X nmkd super scale model is

different versions of that and then also

down here you'll find the 8X nmkd super

scale model so that you can experiment

with these upscales to see which of them

give you the best results now when we

talk about all these different models

and lures and also control that models

the important thing to remember here is

that they have to fit the version that

you want to use so if you want to have a

SD 1.5 model you also need to have an SD

1.5 Lora an SD 1.5 negative embedding

and also for control net you want to

have a SD 1.5 control net model and of

course when you're working with sdxl the

Lora has to be sdxl and the control net

model also has to be sdxl now this is

not true for the vae and this is all

also not true for the upscaler that you

use in high-res upscaling these models

can be used on any model now let's look

at some other interesting facts here you

can see down here I have a tab for

Generation but next to it there is

textual inversion and there's also Laura

textual inversion is the embeddings and

this is mostly used for the negative

embedding so when you click on that you

can see here I have a list of the

negative embeddings that I can use you

want to click here into your negative

prompts field and then click on one of

these negative embeddings and as you can

see here this is added now this does not

use any kind of bracket or any kind of

other information it is just simply the

name of the negative embedding this is

however different for your Laura so now

we're on the Laura tab you want to click

into your positive prompt and now when I

click here for example on ADD detail you

can see that I have the pointy brackets

but also at the end I I have a weight of

1 here now often what you want to do

here is to maybe try this out to lower

it for example to 0.8 or 0.7 as a weight

the reason for that is that often these

lures are over trained so they are

actually too strong to be used and you

want to reduce the weight to get a good

looking result with that also you will

sometimes find that the lower you try to

use does not want to play well with the

model you're using so to combat that to

still get good results there are two

ways you can try to fix that one is to

have higher amount of sampling steps

here than you would usually use so for

example use 40 or 50 steps and another

thing you can try is down here you have

the CFG scale so for that you want to

lower that a little bit for example to

five four or even three to see if this

will improve the image if nothing helps

the Laura just doesn't want to work with

your model also here's a little

explanation for the CFG scale and what

it actually does the CFG scale means how

close the AI is sticking to the prompt

that you are writing so when you have a

lower value of that the AI is taking

more freedom with the image but it can

also often create better results because

the AI can stick closer to what the

model was trained on when you set a too

high value for example 14 15 16 you

might get a lot of problems with the

results in the image and it might not

look good now here's another interesting

thing I want to show you on the right

side you can see we have here these

colorful buttons and one of them when

you Mouse over assess image to image tab

so when you click on that the image and

all of the image information is sent to

this image to image Tab and this image

to image method is actually surprisingly

powerful so here I have the image we

have seen before now I'm changing the

prompt up here instead of writing

beautiful French woman I'm writing

beautiful black African woman and I'm

sitting down here my denoise strength to

0.5 now again for the denoise strength

the lower the value is the closer it

sticks to the original image the higher

the value is the more it looks different

from the original image so with 0.5 it

stays relatively close but also allows

the AI to make changes and as you can

see here we have a beautiful French

woman on the left side and then when I

click here on the image I have as a

result we have a beautiful black African

woman as a result she's wearing the same

clothing it is the same scene with the

same background so that you can do

subtle changes in the image for example

changing the ethnicity but the

composition and the rest of that stays

the same and that can be a very powerful

tool to to adjust your image and play

around with an image you already have

and you like but you want to improve on

that so image to image is really

powerful to get better results another

thing you can do here is to use this

button that looks like a little

painter's palette to send it to the in

painting tab now the in painting tab is

also in the image to image tab but down

here you can see you have different tabs

and one of them is called in paint you

can see again I have the image here so

as you can see on the right side here we

have an icon for a paintbrush I have a

slider here for the size of the brush

and with that for example I can paint

out the face to replace it or I can put

sunglasses on it or I can have an older

person anything I want to do also for

any other part of the image now to make

this work and this is really important

you want to scroll down here and you

want to click here on only mask and then

also you want to adjust here the render

size so often for example in this case I

have something that is almost round so

for the render size I would set for

example 512 by 512 or 768 by 768 because

this is only rendering that part and

then in painting it into the existing

image so don't use too high resolutions

here because it is going to lead to bad

results also another thing here is

because we want to change the image you

need to have the denoise strength high

enough to enable that you want to play

around with that but for example a value

between 0.4 and 0.75 is good with that

to see how soon can I actually change

that so of course the closer the changes

to the original image for example I want

to have the same face but I want to have

a angry expression for the face I can do

with a lower value of the denoise but if

I want to have a more intense change of

the face for example making it the face

of a 70 year old woman you want to have

a higher denoise value to be able for

the AI to actually create that kind of

image now at that point let's go back to

our text to image Tab and we want to

scroll down because I want to show you

some of the most useful extensions you

want to have now one of them we already

talked about is called a detailer and

here for example you can track the face

the hence the person in total and also

more choices for the face now often this

works good here for the face what I can

do here for example is I put a Laura in

here for example here I have a Laura for

Nicki Minaj and then I can just replace

the face with Allura of Nicki Minaj onto

that body now the reasoning for that the

why you want to use this extension is

you want to use the Nicki Minaj Laura in

your main prompt with a low value for

example 0.5 so so that the face is

attempted by the AI but also the body is

fitting the face of the character but

then on top of that you still give

enough freedom for the model that you

are using so it's not overwhelmed by the

Laura so you get the full quality of the

model but then also you get the body

shape and the face shape the head shape

from the Laura now afterwards in the

second step with the a detailer which of

course you need to enable here this is

then rendering only on the face the face

of the character you want to have so

that you have the full quality of the

model on the rest of the image and then

mainly the lower quality on the face of

the person so it looks as much as the

person as possible another important

extension of course is control net with

that you can do all kinds of tracking I

have multiple videos about that that you

should check out you can use as second

extension with that so here you can see

I'm using open pose and this is tracking

the body of the person on the left side

in my preview here but I have another

extension called the open pose editor

installed this gives me this little edit

button here so when I click on that I

have here now a preview of everything

and I can move these points around to

adjust them if I feel like this tracking

hasn't done a good job or I want to

change the pose a little bit or there

are some body parts missing so you can

see down here I have person one I can

scroll down and there are some

informations here like the left hand the

right hand that have not been tracked so

I can add them to the image I can post

them in here and then when I'm happy

with everything I click up here on the

send post to control net button so let's

do some changes here and put this arm up

here like so

and then when I send this you can see

that now this has updated down here and

this can now be used to generate the

image so this is very very useful as an

extra extension another extension you

want to use here is the Roop extension

again I have a video about that here now

what this does is that you can use any

kind of photo and then the AI will use

the information of that face of the

person to render it onto the AI image

now the good thing here is that you only

need a single image to put it onto the

body onto the face of any kind of AI

image that you have rendered the

downside of that of course is that the

body type the head type and the rest of

the person does not match the face or

the body that the person actually has

because it's only replacing the face so

you might want to try to render an AI

image it is as close as possible to the

body of the person and also the head

shape and then use Roop to put the face

on that because otherwise it's gonna

look a little bit strange as a result so

here you can see an example of rupe in

action this is the original image with

the original face of the character and

here we have the same image but now we

have rendered the image of the other

character on top of that the rest of the

image as you can see is completely the

same now here's another useful

information especially for people with

slower gpus older computers when you

send your image to image to image to

upscale it or upscale it even further

what you can do instead of using the

size down here is that you leave this as

the original size you scroll a little

bit lower you go here to script where it

says SD upscale selected from that

pop-down list here and what this is

doing is it is creating tiles and this

is also used using the app scalers that

we have downloaded before words for

example you can see here the 8X nmkd

super scale upscaler this is using a

scale factor of two and what this is

doing is it is splitting it up into

smaller tiles so that you render the

images one after another and afterwards

they are combined into one big image and

last but not least I have another advice

here for you for the text to image tab

but also for the image to image tab when

you scroll down here you have batch

count and batch size this allows you to

render multiple images or image

variations at the same time now what

batch size is doing is that these images

are rendered parallel at the same time

if you have a slow computer and older

GPU that might be a problem for you on

the other hand when you use the batch

count these images going to be rendered

one after another and that should help

you with a slower compute or an older

GPU let me know if I left out any

amazing advice that I should put in

another video like this leave a like if

you've enjoyed this video thanks for

watching and see you soon bye

oh you're still here so uh This is the

End screen there's other stuff you can

watch like this or that's really cool

and yeah I hope I see you soon uh leave

a like if you haven't yet and well um

yeah