Create Easy Ai TEXT EFFECTS in Stable Diffusion // Tutorial – A1111

you can create awesome AI text effects using stable diffusion check out how cool these examples are and by the end of this video you will be able to create similar kind of imagery this is a pretty easy tutorial all you need is two things stable diffusion and control net I've made an entire video about installing stable diffusion so check it out if you don't have it already and the second thing is control net we will be looking in into installing it soon by the way I also wrote an article on this exact same topic on my blog createx a.com if you prefer a reading you can follow it and also check out some extra imagery that I don't show here now if you have control net installed you can skip to the next Tim stamp if not here is how to install control net if it's easier look at the article to follow the steps we'll be installing SD VB UI control net extension for automatic 1111 the GitHub page also includes installation process and some other tips and information first we need to go to the extensions tab select the install from URL Tab and paste this link so here we go to the extension pasting the link and then click the install button wait for the message saying that the extension is installed it can take a few minutes then you can restart automatic 1111 web UI by closing the terminal so you need to completely restart it you could also check for updates in the install Tab and if there are any updates available for control net click apply and restart UI to updated now you should be able to see a new control net tab it's right above the scripts usually and here you can drop your reference image select all the pre-processors that were probably all installed already automatically and a model last time I checked models need to be downloaded separately so we need to go to the models page and download download all the files with extension. pth you can download some of them or all of them but for this tutorial we need depth so make sure to have it installed after you download it drop it into the control Nets extensions models directory stable diffusion web UI extensions SD web UI control net and models as you can see I have a bunch here most models say sd15 somewhere in the name but if you want to use it for sdxl there is also XL models available that you can download and we have depth here as well and there's also things like T2i models that you can be downloading but we really just need depth for this tutorial next up preparing text before you can generate cool looking text effects inside stable diffusion you need to prepare some jpeg or PNG files with the text you want to transform you can create your text file in any photo Pho editing software like Photoshop if you have the subscription or canva canva is free and this is what I'm using right now you can see all these examples of text that I created for this tutorial so create a new empty page in the size you want to generate AI art with I use 1920 x 1080 but any will work just remember the ratio for your later AI Generations type in the word you want to use it can be anything but the more words you include the harder it will will be for stable diffusion to get all the letters correctly as we will see soon and pick a font that will complement your idea for example a handwritten font would work great for ropes noodles wines toothpaste and stuff like that and a blocky thick font will do better with rocks bread ice cubes and so on if you want the text effect to be the main point of an image make it larger and centered or you can also make it smaller in place to the side if you want other elements or portraits inside also then just download all these files as JPEG rpng files and you're ready to begin generating so we're going to go to Sable diffusion select your width and height that are the same ratio as the text files you created I did 1024 by 576 as it's the same ratio as 1920 x 1080 then inside control net we can drop our text make sure to click enable and I like to also select Pixel Perfect right now as you can see there's no difference at all but in the preprocessor we have to select invert and then model depth and when we click to preview the pre-processor results we can see that the text and the background were inverted and this is exactly what we need for this tutorial and all these settings stay basically as they were so now it's just a matter of selecting a seed writing your prompt negative prompt picking your checkpoint and hitting generate and here we have adorable ghosts that say boo so another thing to look at is the ending control step how soon in the process do you want control net to stop affecting the AI generation it could be helpful to play with it so that the results are more unique and don't simply follow the text so the original one was one which means it goes 100% control net is affecting the generation for all these steps when we lower it to 0.5 at about step 12 or 13 out of 25 control net stops affecting the generation and when we run it again you could see that the difference is not huge but there's definitely more changes in the second one compared to the first one and if we go to something like 0.2 look how it starts with a text but then completely disregards it and now we just have adorable ghosts if we do 0.35 it stays for longer period of time but then once again completely disregards the shape so that's something to play around with so when we look at all the results for buo you can see how stable diffusion transforms the text to fit your prompt but keeps the shape of the text image which is pretty cool I also did happy Halloween by adding prompts like pumpkins and bats you can see how right away the text of happy Halloween was prepared for anything that would be thrown at it because it's appropriate for what we're looking for for Year 3000 I did the Highlight option stable diffusion can also work with that and turn it into kind of a light and it turned out all right now for fire check out this font the generation looks like this and I think it's super super cool how it turned out now a font that's thin lines like this thin but large kind of remind me of gates so I decided to do like a gold and Golden Gates or golden Vines and that turned out pretty cool pop this is a perfect font for something round and soft and fluffy or something like bubbles so I decided to go with bubbles for this one and I think they turned out really nice now for this one I wanted to test if we could do a quote and the results were all right with the larger text being mostly visible mostly correct but the Crea x.com at the bottom is all messed up so the size of the text is a bit too small for this generation here is a list of fun prompts to try water fire pretzel noodles gum bubbles rocks gold wood ice neon toothpaste clouds flower I hope that at least one of them interests you and you will test it with your own idea but I still want to emphasize that the text font you select will be very important and will determine whether or not the result works out now let's talk about using multiple control Nets to kind of have more power over the final image so I went to unsplash and found a photo of water that I liked downloaded it and dropped it into the control net unit one and I decided to test it by running the second control net unit at 0.5 control weight and then I'm just copying everything from an image I already generated so I can compare the two after I run them so here was the text it says water and here was the final generation with just one control net now if we use the second one at 0.5 weight we get results like these by adding depth with an ocean reference image we can see that the biggest difference is that of composition but the image itself didn't turn out as well as without the second control net so don't think the more the better you know sometimes that's not it at all but using soft edge control net resulted in a stunning AI generation with a clear composition and lighting I definitely like it better than the original what about you and the same idea applies to all the other ones we have segmentation Shuffle tile instruct P2 2p they all create completely unique compositions and colors and affect the image in all sorts of different ways and I love looking these results they are surprising and they're beautiful you could really use an experiment with all of the control net pre-processors to get the most impressive results I won't be going over what each one of them does or means but probably a series of control n lessons will be coming soon I've been working on collecting data and tests for a while now for control net so if that's something you're interested in make sure to subscribe so that you don't miss it when it finally comes out now another cool thing I wanted to show you is how to in paint text effects let's say we got this image and we really like it but we don't want it to be called a fart shop or nart shop or whatever um so I'm sure there's multiple ways of fixing it but I want to show you how I would approach it so first I would send it to extras and upscale it you don't necessarily need to do that and your life would probably be easier if you don't but this just something that I like to do then I drop that image inside canva or photoshop and don't worry if the image doesn't fit exactly you can still adjust the composition because of what I'm going to be showing you next Once I'm happy with that I like to create a shape and somehow cover the text that I don't like by just selecting a similar color to its outside and that is about it in Photoshop you could do content aware fill or something like that I'm sure you know what I'm talking about if you use Photoshop then we write the text that we want in this case I decided to go with creax AI this is the creax AI shop finally what we need to do is duplicate this entire thing and then delete the photo delete the little background and change the text to Black now we have two images to work with and I'm going to download both of them so I'm going to go back to stable diffusion and sent the original image to inpaint so that all the settings and parameters are the same as previously but I will swap in the new image we just created then inside control net I'm going to drop in the text and the reason why we want to do it this way is so that control net shows St diffusion where exactly the text is supposed to be so it has to be properly aligned with the image do the usual settings and in this case I decided to go with only masked when I hit generate it starts generating and now we have this text but it's way too dark and we can still play with it we could for example decide not to invert so just use none as a pre-processor and run it again and then we have a little bit of a brighter text but it still doesn't look perfect you can play with it by adding other secondary models if you wanted to or you could always select to go for the whole picture and hit generate all right so I upscaled it so it looks like it's going to take a while to generate but basically because it's generating them at the same time in its entirety it's probably going to look better and I think this result is definitely the best out of all that we accomplished and then you can apply the same thing to any other image like let's say a book cover or something you can add whatever text that you want with this same idea by the way if you want the text to blend with their background more effectively try lowering the control net weight like I showed you previously I hope you enjoyed this quick stable diffusion text effects tutorial thank you for watching and hey if you're still here check out this one next cheers

Channel: CreatixAi

Views: 7,922

Keywords: ai text effects, stable diffusion, creatixai, controlnet, tutorial, text effects, create ai effects, controlnet tutorial, controlnet depth, a1111, ai, artificial intelligence, ai text to image, ai design techniques, design with ai, automatic1111, stable diffusion controlnet, generative ai, control net, controlnet automatic1111, stable diffusion tutorial, a1111 tutorial, controlnet 1.1, ai art, text effects in stable diffusion

Id: dtrjVbPEyac

Length: 13min 28sec (808 seconds)

Published: Wed Oct 11 2023