SDXL TURBO - Local Install Guide - 1 Step Render in 0.2 Seconds

Key Takeaways

this is the future of AI hello my friends how are you doing today I want to show you AI image generation that is straight up magical sdxl turbo this is rendering with only one step and creating images in real time while you type and don't forget to enter into the open art comfu ey workflow contest they have a prize pool of over $113,000 with five categor and four special Awards but not only the winners are getting money prizes but also the honorable mentions so there a high chance for you to actually win money for your workflow I'm one of the judges and also on the judges panel is Scott deweer Matteo who created the IP adapter akat suusi and groer both of them are gurus in the comu I workflow scene submitting to the contest is very easy and free so just go to the page I linked below and upload your workflow you can upload multiple times to the same contest so here's the workflow that I want to show you this is based on a video by Scott de Vier so first of all we're going to download the sdxl turbo model I have the link below the video you go here to files and then you scroll down and here you have a sdxl turbo one fp16 safe tenser model so download that either into your comi models folder and in there into the checkpoint folder or if you prefer to load it into your automatic 1111 folder in there into the models folder and in there into the stable diffusion folder with all of your other models next let's have a look here on how to build this workflow I will also link it below the video you can download it and use it right away so we're getting started here with a normal checkpoint loader where we load our model of course and then we need here two text encoders of which we only need the positive prompt there is no use for the negative prompt at the moment so this field here stays empty and then we are not using a k sampler instead we are using sampler custom so when you double click here on the background and type custom you will see here the sampler custom now here of course we are going to connect our model the positive and the negative prompt but then we also have a sampler and the sigas now this is a little bit more special for the sampler this is pretty easy just click and drag out and then you're getting asked and you want to connect this to the case sampler select for the SD tuo scheduler this should technically work the same way but when I drag this out for me this is not suggested so again double click on the background of your canvas and type turbo to get the SD turbo shedu and then connect the sigma output to the sigma input if you don't have the SD turbo schedular note this means your comi is outdated so what you want to do is to click on the manager over here and then click here on update all you want to let it run through to make the updates then close down comi restart comi and let it finish all of the update installs if you don't have the comi manager you need to install that I will link the confi manager below then of course you need an empty latent image that you want to connect here to the latent image input we're going to set that to 512 by 512 now let's talk about the settings you can use any kind of seat you want to use here then for our control after generated we're going to use fixed I'm going to show you why we're going to do that in a second and then for the CFG scale you need to set that to one otherwise it's not going to work of course we need to have a vad decoder and and then here a preview image don't use a save image because this will create hundreds or thousands of images so you don't want to keep them all now here's the reason why we set our seat here to fix you want to go over here to your queue window you want to click on extra options and then on out to queue as soon as you start the queue this is rendering the images as fast as possible so basically in real time and that means when I start here and then I start to type different prompts over here for example at night this is going to update the image right away let's type sunset or we can type for example in London and you can see this is updating right away because this is only using one step it's rendering super fast if you want to stop the que just unhook the auto queue and this is stopping now as you can see from these images the quality of the image is not going to be great because this is just for experimentation it's a research model but of course if you want to you can also build a little bit of an extra here with a loader and the K sampler to render this from image to image to get a higher resolution and higher quality version of an image you enjoy straight up magic my friends let me know in the comments what you think about that leave a like if you enjoyed this video and 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