How to Install and Configure Forge: A New Stable Diffusion Web UI

in today's video I will show you how to install run and configure a new stable diffusion web UI called Forge uh if you

go to this link on GitHub you can find more about this version it's very similar to automatic 1111 but has some

improvements that can be helpful I tested this on another computer uh with 6 GB of video RAM and I noticed a speed

boost compared to the automatic version It's recommended to have 8 GB or more of video RAM but it also works with six

gigabytes to install it you click where it says download and then on your drive create a new folder and give it a name I

will name the folder Forge then save that archive it's quite big so it will take a

while to download on that page it will also give you more instructions like how to run it for the first time and how to

start it you'll find more screenshots and information to read while the archive is downloading once it's

downloaded you can extract the contents of that archive as you can see you have a few folders

including web UI exactly like in the automatic 1111 version plus some extra dobat files to make it easier to run you

can delete the archive to save some space and let's open the update. bat file this will update Forge to the

latest version so you can do that from time to time to make sure you have the latest version press enter to continue

now let's open the run.bat file this will start the forge interface after installing all the necessary extra

things that need to run uh the good part of Forge is that it recognizes your video card and auto adjusts uh to the

best settings so you don't have to play with all kinds of settings until you get it right so this part takes quite a long

time so you need patience from here I can see it installed all it needs for the interface and now it needs a stable

diffusion checkpoint model to run as you can see it downloads the realistic Vision model and shows the folder where

it's placed it stayed on 100% for quite a long time so I decided to check to see if it's complete um I go to where all

the models for stable diffusion are in the web UI folder then models then stable diffusion and as you can see here

it says partial maybe I didn't wait long enough but this is a good occasion to show you how to install your own models

so what I will do now is close that command window and delete that partially downloaded model then go to civit AI com

website you can click on models to see all the models um from the filters you can

filter them by the most downloaded in the time period you want um select the checkpoints and the most popular ones

are sdxl and the SD version 1.5 my favorite which has a new version recently released is Juggernaut XL you

can see it appears here but if not you can go and search for it download the latest version which is currently

Version 9 just released then click on the big download button and as you can see it's over 6 gbt now an important

thing to note when you download checkpoints make sure the extension is safe tenser now put it in the web UI

folder then models and look for stable diffusion click save and take a break again since it will take a while

depending on your internet speed while it's downloading each model has some recommended settings so you can check

those out as you can see here uh there's more info about the model and what settings

are recommended after it's downloaded we can go back and run the run.bat file again to start forge while

it's loading you should know that this installation includes everything it needs and doesn't interfere with other

installations of stable diffusion if you're not happy with it you can simply delete the entire folder and as you can

see it started now how it should for those who have the realistic Vision model downloaded completely you can go

and test it out but for me since I downloaded a new model I will go to where it says

checkpoints and here you should see all your models I will choose Juggernaut if you downloaded a model while stable

diffusion was open you can just hit the blue refresh button and then search for it in the list now let's try it out

quickly with a prompt like a photo of a cat remember to look for the settings of your downloaded model model for sampler

steps and CFG scale samplings seem to be good so I will just add some 30 sampling steps for width and height the

512 pixels is for version 1.5 models for sdxl models you want something like 1,24 pixels and uh for CFG scale between

three and 7 uh let's hit generate and see how it works um it took almost 7 seconds on my

computer now let's close stable diffusion both the browser interface and the command window let's open the

run.bat file with notepad uh you can use edit with notepad uh or notepad++ if you have it installed as you can see here

it's calling the folder web UI and there is calling another bat file named web uiuser dobat so let's go to that folder

and search for that file once you find the file you can open it again with Notepad here we have the command

arguments as I told you earlier Forge does most of the settings automatically but if you still need some extra

arguments here's where you put them I will put line line theme and dark then save and close the

file and go back to the original run and see how it works as you can see it's launching with that dark theme argument

and it starts with a cool Dark theme have you noticed that uh after starting again the sizes Samplers and anything

else are resetting back to default each time you start stable diffusion um isn't that annoying so

let's put the settings back and make sure they stay this time to do that go to the settings tab there on the left

scroll down and look for others and there you have defaults click on it and scroll back up if you click on

view changes you can see the old values and new values click apply apply settings and

then the reload UI button after reloading all should remain with your settings just keep in mind it takes

everything you have in that moment in the interface including prompts let's talk about extensions if you go to the

extensions tab you can see all the extensions that Forge comes with by default these are like plugins that

expand the stable diffusion capability click on available and then then click on load from now you can see all the

extensions available here are two extensions that I use all the time search for sdxl and style

selector this will allow you to test different styles click on the install button and wait for it to install Now

search for rtio helper this will let you change the Rao easily install this one as

well now go to the installed tab if you hit check for updates you can see if there are updates for existing

extensions then press apply and restart button let's see where those uh extensions appear if you scroll down you

can find sdxl Styles it's enabled by default but if it's set on base that it doesn't do anything so you can have it

always on now here we have the ratio helper you can choose a ratio that you like

and then you can adjust a slider for size and it will keep that ratio if you use those arrows you can switch from

portrait to landscape you can also turn it off and let's say you have a different ratio you can put your pixel

size in there and then select the lockpad icon now it will keep that ratio it's quite helpful in my

opinion let's select a ratio and scroll down and check the style and we will do a test on base when it doesn't do

anything let's add a prompt like a cute cat and hit generate now see where it says seed and

shows minus one that is the random seed you can set that to random anytime from the dice

icon but if you use the green arrows recycle icon it will take the last seed which in this case is the seed of the

cat if it will be to generate again it will get exactly the same image but I want to show you how the the style

influences the result on the same seed select a style for example impressionist hit generate and you will get a cat in

an impressionist art style if you look under the image you can see what it does it adds before and

after your prompt some extra words to define a style and also it adds a negative prompt let's try another style

maybe the fantasy art one now if you want a bigger size you can enable highers fix

I will choose dx2 upscaler but you can choose others that fit better your generation for upscale I will upscale it

two times but if your stable diffusion crashes you can see that in your command window you can restart stable diffusion

and use 1.5 upscale instead of two for denoy strength I play around between 0.2 to 0.5 depending on how similar I want

it to be with the original image hit generate and wait for it to finish maybe this seed in style wasn't the best

example for upscale but let's try another one I usually create a lot of versions without upscale and then use

that highest fix on the version I like I set the seed to random and generate again now let's try another style like

anime so I generate a few times then when I like something I stop I like this one for

example now let's activate H's fix and see how it will look if we double the size oh no it's different why I

intentionally didn't lock the seed number to show you something so uh I like the previous one

but how can I redo that version open the folder uh where your images are generated uh and as you can see here the

previous image is there but it has a number and the prompt I will copy that seed number and we'll add it in the seed

field now I can get that exact image upscaled now look at all these details in the fur in the carpet below it does a

nice job if you play around to find a good seed so let me show you again how I do it I uncheck the higher fix and I set

the seed to random I generate until I like something on the image I like I stop and reuse the seed then I enable

hires fix um let's go to settings so I can show you some settings um I use um for

the path for saving uh I prefer to have allinone folder but you can copy and paste the path of where you want your

generated images to be saved and then for saving images you can choose the format and the image file pattern um as

you can see now I have a number a seed number and a prompt if you click on Wiki you can find

all the names there and you can configure how you want the file name to look so I will delete what I have and

reconfigure I will put the seed there first in case I need it then for prompts you have different options depending on

how how you want it to look um I will go with prompt words so I will copy and paste that one as well now here you see

it says it adds a number when it is saved you can also replace the existing image if it has the same prompt and seed

or you can add a suffix I will go with ADD suffix and I will uncheck number to the file

name press apply settings and then the reload UI button I will test the new settings really quickly with a prompt of

uh a cat in the snow for the style um I will go with digital art and let's check what file name we have now um as you can

see we have the seed and the prompt a clean file name let's reuse the seed and see what happens when we have exactly

the same impr prompt and Seed as you can see it adds a suffix number like in this case one so I think this will work for

me so I don't accidentally replace the images let's go again to settings to see what else we have there for saving to

directory by default it makes a new folder with the date but I prefer to have all in one place and clean up each

date and sort it myself if you scroll down you can also have settings in UI if you click on

info you can find a list of all settings you have settings that appear on text to image and for image to

image um that means those settings will appear here in these tabs Let's test one to see if it adds it I will add the word

tiling to the text to image apply settings and hit reload UI as you can see it added the tiling here

I will check a quick quick prompt without it like some red roses maybe in a impressionist art style uh if you open

this image and click on preview tiling we can see how it's repeating you can see that it has this line in the Middle

where you can see it's not repeating perfectly let's reuse the seat of the image and see if there is any difference

with tiling activated after generation we check again and surprise it doesn't work in

the automatic 1111 version it works works fine so I checked the forge page on the issues Tab and found that a user

already posted this as a bug so for now the tiling doesn't work as it should and let's hope they will fix it soon you can

try to update Forge in a few days maybe they can fix it you can also add more things in this area at the top of the

interface if you go to settings then scroll down to user interface you can see here on the Quick Settings list all

the items that appear on top like the the checkpoint and so on here you can click and add the element you need click

on info for more information about what elements you can add in the left is the name and in the right is the description

when you are done click apply settings and the reload button let's explore some buttons as well with this one you can

reload generation parameters from the last generation as you can see it added all

the prompts seed and settings the next button clears The Prompt and the last button applies selected Styles

but I don't have any Styles saved so let's save one let's say you have a prompt and negative prompt that works

well for you and you use it all the time click on this brush icon and you have some buttons for style options click on

the last one to copy the prompts from the interface to make it General leave in the prompt only the Styles not the

objects unless you need those to appear in every generation then give it a name press the

save button to save it um all the save Styles appear in this list let's clear the prompt first and show you how it

works I will select that style and now I will add a prompt with a cat after I hit generate it will create a cat in that

art style if you look under the image you can see how it added the actual text from the Saed style after my prompt and

also added the negative prompt if you have more Styles saved you can select multiple styles from the list now I can

show you this button that applies the selected style to the prompt as you can see it added that text from the style to

the prompt and cleared the style from the list so you can edit the prompts more before the Next Generation if you

go back to Styles you can select the style from the list and edit the text how you want it and save it or you can

press the delete button and it will permanently delete that style if we go to image to image you have a new button

that looks like a paperclip called interrogate clip if we upload an image here like let's say this bunny I can

interrogate that image basically it's a model that can reverse the image generation process and give you a text

prompt for it uh just the first time you use it um it will take some time to download you can check that in the

command window uh but after that you will see the prompt appearing now you can use that prompt in text to image or

here on image to image to get variations of this bunny I will hit generate and see what we get and we got a distorted

image that happens because we had a different ratio we generated with a square ratio but the input image was

portrait you can change the size manually or we can use the ratio helper extension again and click on this

triangle to get the image size now if we test again with the right size it should work correctly you can play around with

deno's strength the bigger the strength the more different the variation will look we can also change the style in the

process like maybe I want a 3D render look um I will choose that from the list and try again looks quite nice and clean

we have more buttons here here are some useful ones you can send it to image to image so you can get more variation on

that or you can send it to extras to get more modification on it here I can for example do an upscale of it I will

choose two for resize and an upscaler like the DAT when I hit generate I will get a bigger version of

it as you can see I have doubled the size of the image let's go back to text to image and

see what else we have here there is something called free you integrated if you search that on Google you can find

more info about it on GitHub it says it improves the quality of the image at no cost and no increase in time if you

scroll down it has different settings depending on the model for sdxl it has these settings so you just put the

values for B1 B2 S1 and S2 there I will give it a test with a cute dog to see if it offers what it says so this is what I

got now let's see how the results are different with the option activated I will keep the same seed and prompt and

um I will just enable free you they said it will Notre inre the time but on my computer it takes longer than without it

so I'm not sure if it's because of Forge or something else the results are definitely different but I guess it

depends on the model and what you are after in the left is the original and in the right the free U version it seems

more clear but also more cartoonish and more colored uh so if you get a nice generation but it has some errors you

can try free you to see if it can fix it for you let's see what else else we have here it says control net integrated if

you click on that title it lets you add multiple control net units I found that by mistake enable control net then let's

drop an image here maybe this cute dog I will enable Pixel Perfect and then for control type um let's go with canny we

have a pre-processor but for the model it says None So what do we do uh we go online

and find one um on this page there are a lot of models for S DXL version a I will add a link in the comments later if I

don't forget um one that I tested on uh automatic was this Coya control light XL canny it has a download button and let's

see in which folder we can put it so Forge can recognize it I will go to the web UI folder then in automatic it was

in um extensions but since here it says integrated it must be somewhere else look for for the models folder then for

control net and there you can save it now if we go back to forge and look for the blue refresh button uh click on

that to refresh all the models in the list and um as you can see now the model is in the list I will test it quickly

with a cat prompt and we got this wannabe dog cat we can see next to it how the canny processed the image and

what Contours it got if I don't want to be so strict I can reduce the control weight so now it kind of kept the

composition but got a nice cat instead you have more models that you can play around with maybe you can try

some for depth or open pose or IP adapter if you have questions you can join my free Facebook group called

pixaroma Community we have daily AI challenges discussions about AI design and crafts I check the group daily so

that can be a place for you to connect with other users and me of course on YouTube I add new content weekly so take

the time to hit subscribe or like the video that will help me a lot I also have some recommendations for other

YouTube channels that can assist you in your AI Journey one of them is Olivio sas's Channel I hope I pronounced that

right uh the second one is nerdy rodent which has some interesting tutorials and another one that I enjoy is the atre

preneur channel obviously all are more popular than mine but I am getting there step by step I am almost meeting all the

requirements for monetization I just need a few more hours and I I want to thank all the people who helped me get

this far until next time have fun with AI and see you soon with a new tutorial