Best NEW AI tool? InvokeAI tutorial for Stable diffusion.

Key Takeaways

Hello friends. Will this new UI change your life when it comes to stable fusion? Maybe. Is automatic 111 king in your opinion? It is for many people. Today we will be looking at a contender that has great potential in dethroning automatic 111 over time. In this tutorial, we'll install it and check it out. Let's get started. Okay, so here we are at Invokes GitHub page and what you need to do to install this is just scroll down a little bit and this is super simple. They have a installation quick start here. We're going to ignore all of this for now. Just going to go to the latest release page. Going to scroll down to the bottom and here you can find the version for your OS. It has support for Linux, Mac, and Windows. We're going to download the Windows one. We're going to take this and we're going to drop that into a folder that I've created called Invoke AI. And then we're just going to push this install button here. Now it says it's protected your PC. This is all fine. You can just press more info and run anyway. And it says here some of the installation steps take a long time to run. Please be patient. If the script appears to hang for more than 10 minutes, please interrupt with Ctrl + C and retry. So, press any key to continue. Now, it's downloading some files and I'm just going to speed past this bit. So, uh I'll see you when it's finished. While this is installing, I thought we're going to talk over a couple of requirements. So, as we said, this fork is supported across Linux, Windows, and Macintosh. And if you're on Linux, you can use both an Nvidia card or an AMD. But for Windows and Mac OS right now, you need an Nvidia card with at least 4 GB of VRAM or an Apple computer with an M1 chip. If you don't have this, you're going to have a bad time trying to run this. It says here they don't recommend the GT GTX 1650 or the 1660 series video cards as they unable to run in half precision mode and do not have sufficient VRAM to render 512x 512. If you are on these cards, I mean you could try obviously, but uh automatic probably works better for you then. You're going to need about 12 gigs of uh disk space for the machine learning model and um Python and all its dependencies. But these are all going to be automatically installed. Okay, so the installation has finished and this took roughly 10 minutes actually. And then it asks here where your models configuration files will be placed and image outputs. So just accept these locations. Then it asks, do you want the not safe for work checker? This is the sampler to use by default. Now it sets K hune as default. I want I don't want that one since it's fairly slow. We're going to go with Kuler A as our default. The number of D noising steps. Default, we're just going to leave that 20. And here's a step if you want to download the models. And if you want to download the recommended models, it's about 10 GB. Or you can skip this step or and put in your own models. If you're doing this for the first time and you don't have any other UI, I would recommend just going with default, which is the recommended models. And just for the sake of this tutorial, I'm going to do that as well. So, I'm just going to press enter here. And it's going to download models on about 10 GB. And it says here license agreement for weight files. And if you haven't done this previously, you need to go into hugging face and you need to create an account which I have here. And then I'm going to link the files, the models, which is the stable diffusion 1.5 and the stabilusion in painting license agreements. So it would look something like this. You need to agree to share your contact information to accept this model. And you're just going to press I have read this agree. You should do that with both the 1.5 and the stabil fusion in painting. Now stabilusion 2.0 is not yet available here but it will be soon and just press enter. Then go back into your hugging face account. Go into settings. Go to access tokens. Gonna create a new token here. It's going to invoke AI. Set this to write generate a token. And then we just copy this with this little button here and paste that right in. And if you're on Windows, you can just right click here and press enter. Now it starts to downloading the weights that you will need. So, this is probably the most complicated step, but if you just follow what I did, just go into hugging face, create an account, go to the two links that you can see in the command line, it's the stable diffusion 1.5 and the inpainting model. On both of these, click accept. And then you need to create your token, which needs to be a right token. and you copy that and you paste that into the command line and then it starts working and downloading and installing the models for you. And after this step, you're going to be completely finished with the installation. [Music] Now, the download of the models has been finished and it took about 3 million years. So, you need to be prepared for that. Now speaking of installations, I once worked as a repairman and um I installed a motor for moving stairway that was um it was a bit too powerful. After that things escalated very quickly. All right, so it says here invoke AI installed successfully. You can now start generating images by double clicking the invoke.bat. So we're just going to press any key to continue. And that goes away. So this is my folder invoke AI and we need to find the invoke.bat and that's the one here. Okay. So we're going to click this and it says you can have you can generate image using the command line the browserbased UI or open the developer console. And what you want to do here is you want to use the browserbased UI. So you're just going to press two enter. And now that starts up. Okay. Okay, so it is now started and it you can find it by going to your browser and entering localhost at port 9090. So we're just going to go do that. You can either do it like they say here 127.0.0.0.1 the we enter localhost colon 1990 which is much easier. So here we have invoke AI. And now the first thing we're going to do is going to press this here or just hover it. And we're going to change this to dark. So at the left side here, this is where you have your different options to use while rendering. And what we can see quite clearly here in Invoke AI is that they have worked with the user interface and the user experience much much more than any other UI. so far and I talked a little bit to the developers of Invoke AI and they say they are focusing on the user experience but also on the stability of this. I asked about comparing the features to other UIs like automatic and they say we don't want to implement any features that can't run stable inside our UI. So they want stability, they want ease of use and they want users to have a good time and I think that's a valiant effort while creating a user interface. I know many of us while using automatic 111 have had lots of issues and we don't want to spend time looking at GitHub finding the bugs and issues and trying to work our way around. So I think the stability here is a great factor. So first text to image. This is where you're going to create any image. I'm just going to copy paste something. I have huge colossal statue in Nordic woods and fjord concept art station rocky cinematic underexposed. We can choose how many images we want, how many steps, which is basically how many iterations for the AI to take before finishing the image. and the scale, which is how closely the AI is going to listen to your prompt. So, the higher the scale, the more closely we're going to be to the prompt. So, we're going to leave this at the default 7.5 at now. This is the resolution. We're just going to leave that for now. Now, it didn't give me the uler a as a default for some reason. So, we're just going to run with KMS for now. You can set the seed, and it's set to randomize as default. you can have some variation in your seed and it says here for everything here there's a little question mark so you can read more about what everything is so that's very nice we have the phase restoration here you can choose now it's the same as in automatic 111 default you have the GFP GAN and the code former I prefer the code former so I usually set that now we don't have any face restoration going on with this prompt so I'm not going to activate it for Now upscaling. Now we don't have any options but we have an upscaler. We can do it two to four times. And it says here it's using ESR jan which is a good default upscaler. We have some tiling and we have high res optimization. See now we don't get any settings for that. Okay let's generate this just as a test. They say invoke which is quite clever because they're calling it invoke. And as a default, you have the live render preview. So that's very nice. And it's less glitchy than the one in automatic 11-11 where you can see steps jumping back and forth. Here you can actually see a more smooth transition. So I think this is quite nice. While you have your image created, you have some options. You can send it to image to image. We can send it to unified canvas, which is the outpainting tool in Invoke AI. I think this is one of the better options and uh we're going to look at that in a bit. You can copy link which is basically just a Ctrl C and you can download the image. Now this is pretty cool. It's a huge colossus of the Hulk like you can open in viewer which is basically just you know looking a little more in detail. Now if you have an an image that and you don't have let's say remove this you can press use prompt and it gets the prompt or the seed that you have from a previous image. I can take like this image and say oh I want that seed. I'm going to put that in there or okay I want this seed. We don't want that for now. That's a good feature to have and you can use all the settings from a previous image. Here's the face restoration. And again, you can choose code format give began or you can just restore faces. Click the little button here afterwards. And here you have your upscaling options. We're going to upscale this two times just for the sake of this video. I'm going to press upscale here. And now the processing is complete. So that was a quick look at the text to image which is the most basic setting for creating images. If you have various models you want to use, you can just select them from the drop down here. We looked at the themes here earlier. We have the hot keys here. You can report the bugs just inside this tool without having to go to GitHub. You can find the GitHub link here if you want. And you have a link to their Discord. And if you have any questions about this, you're free to join my Discord, which can be found in the links below. You can ask your questions in the help channel or me or another user will try to answer it for you. And if you want to talk to the developers, you can go to their discord which is here. You have some settings here to the right. This is the in progress images or the live runner preview as I like to call it. If you have any bugs with this, you can reset the web UI here. So the next tab here is the image to image. As we're moving to this, we actually have our previous image here, which is quite which is quite handy. Now, we don't need it, but it's a good feature. Now, you can also take one of the images here and drag it here. And this is the dnoising strength or what they call the image to image strength. And the higher you set this, the more the image is going to change. So, we're going to try this. We have our prompt from before. So much of the composition is the same. We have a better statue in the middle here. And again, you have the same settings to the left here that you have in the text to image tab. Next, we have the unified canvas. Now, this is the big feature here, unified canvas. Now, we're going to take our previous image here, and what we can do here is that we can outpaint. And now looking at this user interface, it's much smoother than anything I've seen previously. You can easily scroll in and out or zoom in and out with the scroll. If you hold down space, you can move around just like in Photoshop or any other tool. We're going to pick this move tool here, shortcut V. And then we can either just drag around. We can also select. If you see when I hover over the image here, you can move the little box here. So that's what we want to do. And we want to keep parts of the image as the it wants to understand what's going on. If I just outpaint over here, it's just going to create a new image that's not close to the first one. Now, because of the prompt that we have, they're similar in color, but it's not what we're looking for. So we're just going to undo that. And then we're going to go back like this. I'm going to try again. Now it's saving this part and it's it's going to add to this part. Now this was terrible. I have no idea what happened here. We still have this part. It's pretty cool. I think this it tried to turn this into some kind of colossus. So, we're going to actually remove the Colossus part. We're going to undo. And then we're just going to keep the Nordic woods and we're going to invoke again and see what happens. And there we go. Now we're closer to what we wanted. Now the lighting is a little off. It's much lighter here, but it's also tighter between the trees and the leaves here. So, I mean that could be accurate. So, just leave that for now. We're going to outpaint to this area. But first, you need to either accept this or discard it. We're going to accept this for now. Then, I'm going to show you another feature where you can do multiple images. We're going to create two here and then invoke again. Now, coloning invoke is a smart move because if people start saying invoke, they're going to remember that this is the tool that they used. So, if that catches on, this might be a front runner. We'll see. Now, you can go previous or next here to see the different images or just press left and right on your keyboard. So, first one, second one. We have some blur here, but apart from that, I think this is pretty cool because you're keeping the same light coming in. This one's also very nice. We're just going to take the second one here. Now, you have some sort of seaming issues, but that can be corrected. Let's going to first accept this. And there are some options here. Seam correction. So, here you can play around with seam size, seam blur, the strength, and the steps. We're not going to go into detail on those here. So, play around with that and see what values work for you. And different values are going to be good or bad for different images. So, there's no perfect setting here. You also have infill and scaling. Now, this is basically the resolution of your inpainted parts. You can start by using this at auto, but you can set it manually. And you might need to up this value if you're doing smaller. Let's say you're doing like a smaller piece here. You're going to move that up here. Let's try and continue this tree. Now, the bounding box here is quite small. So, you need to make sure that the scale resolution infill isn't too low. If you have this at auto and it starts getting blurry, you can change it. I'm going to run it at auto for now. And this is setting the size of this box. See? And also you can see that while you're changing this, the scaled resolution changes here. So this is the box and this is the resolution for the actual inpainting. Now to me, one of the most important features are here are the undo and the redo because they're super easy. I mean, if I just play around with this, I can actually I can delete here. I can go here. I can start painting stuff in with the brush instead of just dropping it back and forth into Photoshop or Photo Pay or Affinity or whatever you're using. And that's to me much more powerful than anything in Automatic 111 user experience-wise. Now, Automatic 11-11, yes, it has more features. You have all the extensions, but this is a new tool. the user experience is much better and I think in time unless automatic 1111 shapes up the user experience it might fall behind. So we'll have to see what happens. Might be some extensions working on improving the user experience. I would love to see that. But for now I have big hopes for what invoke AI is going to do. Now the next parts here are nodes which is not developed yet which is nodebased system for generation of images is says here is under development. Stay tuned for future updates about this amazing feature and you have the post-processing but it says invoke AI offers a wide variety of post-processing features. So image upscaling and f face restoration those we could find in the previous tabs and we're going to see what more post-processing features are going to be available. You have training here. So this is where you will be able to do textual inversion and the dream booth. So yeah I think that was all for this video and I think again we want to talk about the user experience. That's the most important part. the stability that they claim is superior to the other UIs for now and the outpainting. It's part of the user experience, but it works great and it's just it's fun to use. I've been playing around with it and um you try too. And if you have issues, come ask me in the Discord or check out Invokes Discord. Either way is fine. I hope you had fun while um while we checked out this new tool. Do you think it can be able to compete with automatic 11-1ven? Automatic 11 is much more featurerich. Invoke AI promises stability. Do you think it will be a contender? Put it in the comments below. See what people think. Have a good one. See youa.