Stable Diffusion 3 - How to use it today! Easy Guide for ComfyUI

stabil Fusion 3 is here and I'm going to show you how to run it hello my friends how are you doing do you have a tent in your pants because of the stabil fusion 3 announcement well today I'm going to show you how to get your greasy little fingers on it but before we do that I'm going to show you some comparisons between mid Journey SXL and stable Fusion 3 let's get started so here we are in mid Journey because I want to show you that all of these are first roles now here we have something for the ability of imagination so the prompt is just sci-fi movie scene and these are the images that mney is creating and as you can see they are very cinematic and beautiful as you can see we are in comu ey because comu ey always gets the fun stuff first and up here I have my note for stabil Fusion 3 and then save image and below here I have my setup for real Vis version 4 the same resolution as above and here we will see our SXL result stable f iion three very nice very good work from the colors from the composition the guy standing here on the side a little bit awkward from the composition but otherwise actually a really good image and below that we have here our real whis sdxl model also very nice result here interesting composition but it has this kind of classic art style that you expect from stable diffusion Community trained model so and one of the promis for stable diffusion 3 was that this is closer to the aesthetic and artfulness of mid journey and I can really see that here so here we have the second scene that has been generated by stable diffusion 3 again I would say this comes pretty close to the aesthetic of mid Journey the colors are very nice and recognize here that we have only two colors we have orange and red tones here and then we have blue tone so this two color rule is really very nicely followed here and I feel like the interaction between these two characters is also very nice so this is kind of a scene where you want to see what's happening next and this is the result from sdxl now that is very similar to what we have seen before let's go to the next image here this is one of my favorites it's a wolf sitting in the sunset and it's very beautiful very nice result very Artful created by mid journey mitun is really good at these Artful compositions now here the result of stabil diffusion 3 is pretty nice even though the neck is way too long the composition is a little bit awkward but because we have a 16x9 format sa Fusion might have a little bit of a problem with this wider format here we have the sdxl version of that actually pretty good very nice wolves created here this is more photographic this might be down to the model being more of a photo model but the result is nice the colors are nice composition is nice here we have a second render of that prompt with stable diffusion three it's a pretty nice composition even though the character is too much in the center here we have a dark character on a bright background we have a nice highlight at the edges here to point out the edges and also show the hair of the character so that's pretty nice the sun is in the background even though the sun here is not completely round and this again is the result by sdxl the real V model pretty cool again very nice wolves let's go here to the next images and this is actually with text so here we have a tiger that is supposed to be in a pixel style in this case it didn't really pick up on that it is supposed to say I love you so much so that actually worked really well in this image wow so this is the result from stable diffusion 3 I think that is a great result it looks a lot more like a Pixel Character even though as far as I know it's not trained on Pixel images they try to keep their model cleaner now remove artistic references and artist life artist style references but still you can see here a beautiful result the text is completely correct even with the words on top of each other so that is actually a very nice very surprising result here we have the result from sdxl now sdxl is known for not being that great with text and the tiger is pretty beautiful the style of the tiger the design of the tiger everything he is very nice the way he looks the smile he has on his face is good here we have a poodle that is on a fashion shoot and it really looks very good I just want to point out that in this case we even have a much longer prompt so this reads in front view portrait of a poodle wearing 1960s mid-century Space Age fashion backdrop is a deep neutral blue setting a dark contrast that accentuates the subject florescent color well actually you should put a color here shade wearing a very fashionable Lounge coat and sungl glasses are a hip 1960 style clothing all in one color here we have the result from sdxl and oh boy is that result good from the composition from the color from the style from these two different lights are really love the idea here that we have a pink highlight here on the side of the hair very nice and then contrasted with blue on the right side really works very beautiful and here we have the result from the real visz s L model again I would say it is very beautiful really nice and detailed in this case actually I like the hair better now you have to decide for yourself which of these you find more Artful and expressive but as a studio photography goes I have to say both of them are very very beautiful here's a second result and oh man I love this so much it looks so cool very photogenic very nice very beautiful again as you can see here if you look close the hair is the fur is not specifically nice and detailed also up here is kind of fuses into each other but other than that it is just a very very nice image creation now let's have a look down at SXL and wow this is also very good I love this kind of silky material from the dog check and I love the reflection in the glasses the attitude of the dog and this cool vintage decor with the chair with the this kind of circle behind that very nice I have to say like both of them are really really good now let's do something that might be a little bit more challenging for both of these models here we have character emotional expressions of these cartoonish cats let's zoom in here a little bit you can see they look very beautiful this is mid journey of course the prompt here is clip art cartoon cat wearing glasses with a series of Expressions happy curious angry sad tired embarrassed cute and there we have the result from stable diffusion 3 first roll a lot of errors in the face uh the expressions of the face are pretty nice they have Variety in them we actually get six different characters that look pretty similar to each other so it's good and not good at the same time and here we have the sdxl result from that we have seven very beautiful cats here but none of them have any face expression which is really surprising so let's give this another shot here we have the second roll again the expressiveness of the faces is really nice I guess that stable diffusion 3 here needs some more information in the prompt a little bit more details of what we want to create here but overall I feel like the result is very nice there's a lot of similarity between the characters and the expression for example here this kind of angry face down there is very nice here also a little bit of annoy face I would say that is very very nice and for the second image for SXL I actually switched over to chuggernaut but we're getting the same result there is no emotions even though the cats look very beautiful now let's do something that's a bit more complex again here the mid Journey results girls with big guns now that looks amazing love the design of the character of the gun of the composition everything here is just really nice in this case The Prompt is color anime cute pigtails girl with guns a large automatic gun in a bond pose there we have our first result from stable diffusion there we have our first result from stable diffusion 3 now this I'm not happy with to be honest uh it's almost there but the hand is going to the wrong side we might get a better result in the second role here we have the result from chugun out model and really actually pretty nice it has this classic stab Fusion style for the community trained models but from the colors and from the character actually good even though the hands didn't 100% work out in this case I adjusted the prompt now it says color anime drawing of a cute pigtails girl with automatic rifle a large automatic gun in a dynamic pose in a war zone highly detailed Masterpiece for both of the models now this is the result we get from State Fusion 3 and I feel like it's just not picking up on the highly detailed anime style so maybe anime is just not the right thing for this model and here we have our result from sdxl chuggernaut now here again we don't really have an anime style but the image is very detailed the pose is very nice we actually have the war zone in the background so this is actually not too bad as a result let's try another prompt here this is a result that I really enjoy in MIT any these closeups here beautiful color I like this kind of glitter effect on the face if you're curious The Prompt is raw photo of a beautiful French woman style girl with gold eyes wearing sparkly pop Kira Kira rabbit Rings quirky cartoon art style line drawing intense colors Pop Style lines Rich details doll core right pink and right white here we have the result from stable diffusion 3 I would say that result is pretty beautiful very nice face details the hands here a little bit deformed in all kinds of ways but overall the colors and the composition is very nice and this is the result from chuggernaut sdxl which is a pretty beautiful result even though this went more for the photographic style here we have a second roll with stable diffusion 3 again I would say the result is pretty beautiful and here we have another result from chut sdxl again very beautiful result let's also try the famous prompt from the announcement of stable diffusion 3 with the wizard on the hill and the tax buff now here we have the result that I'm getting from stable diffusion 3 and actually it is really working we have the text here the text is well almost correct there's a little bit of an error here but you can always reroll the threes in here we have the magic we have the wizard we have the mountain so all of that really actually works here we have the result of SX L chuggernaut which is really beautiful but it doesn't have the text in there at all and here we have the results from mid Journey now if you look closely down here it says stable Fusion you can't really read it but down in the corner we have some text here the image itself is very beautiful but it is lacking the wizard spell casting and and also the text okay so let's have a look at how to actually install this so you can run it now first of all what this does is it's using the stability API and that means you need to have an account with stability you go to the website you make an account and then you can have here an API created so when you go to your profile on the left side here it says API Keys just click here on create API key and as you can see it will create that key for you now when you first sign up to the website you get 23 free credits this is not for 23 images you can actually go here to the pricing page you can see in the generate list for stable Fusion 3 it costs 6.5 credits per image and then for the stable diffusion 3 Turbo model it costs four credits per image and if you look a little bit lower we have here sdx L1 which costs way pay less than that 0.2 to 0.6 that is much much less than you would pay for stable diffusion three you also have to go to the billing page to buy your credits now the minimum amount of credits you can buy is 1,000 and it cost you $10 to do so if you're in Europe you're also going to pay 20% W or tax on that on top of it so $12 now for the installation it is actually pretty straightforward but when you first come to the GitHub page you will realize that a lot of the information here is actually in Chinese so the way to fix this is right click and then simply select translate to English after this is done you can read the text completely in English and it is very easy to do you need to copy this here git clone and then the address of the GitHub project you have to go you have to go to your come VI I folder in there to the custom notes folder open it up click in the address bar and type CMD and hit enter when you have the command window open simply copy in the git clone and the web address and hit enter so this is downloading the project folder for you after this has been done you want to go into that folder which is called comu I stable diffusion 3 API and in there you're going to find a config Chason file in here you find this text and this is where your I key goes now there you have to keep the quotation mark so only replace the text in here that's very important then save the file and restart comi inside of confu ey there is actually only one note that you need it is called stable diffusion 3 you have an image input you have an image output now this is intended to be rendered also with image to image rendering this is why you have the D noise down here as far as I know for now the image to image rendering doesn't really work to add this note you want to to double click on the background this gives you this search bar and here simply type stable diffusion this is going to show you the stable diffusion 3 Note click on that and this is going to be added for you and then simply you connect this to a save image node now inside of the Noe the settings are as easy as they can be you have your positive prompt you have your negative prompt you have your aspect ratio you have the mode text to image or image to image you have your model which is sd3 or sd3 tour you have the seat you have your control after generated which means you have your seat setting which you can set to randomized fixed increment or decrement just as usual and then you have the strength which usually you want to keep at one and only if you want to go for image to image this would go to a lower value than that to take the image into account let me know in the comments what you think about the stable diffusion 3 modeles leave a like subscribe to my channel if you want to see more videos like that thanks for watching bye oh you're still here so uh This is the End screen there's other stuff you can watch like this or that really cool and yeah I hope I see you soon uh leave a like if you haven't yet and well um yeah

Channel: Olivio Sarikas

Views: 48,711

Keywords: oliviosarikas, olivio sarikas, olivio tutorials, stable diffusion 3, SD3, SDXL 3, Stable Diffusion local, Stable Diffusion 3 local, sd3, stable diffusion, stable diffusion tutorial, ai news, comfyui, comfyui stabdle diffusion 3, sd 3 comfyui

Id: SBkLY703Xs4

Length: 16min 13sec (973 seconds)

Published: Thu Apr 18 2024