SDXL IMAGES IN UNDER 2 SECONDS!! || Comfyui LCM full setup guide

hello everyone today I'm going to show you how to use LCM lauras LCM lauras allow you to make images in SD Exel in under 2 seconds and the quality is pretty dang good um in this tutorial I'm going to be using reality check Excel which is the model I made um under team synerject I will be making more as the XM models in the future but for now uh I do recommend using this model as it gave me the best results promise I'm not biased uh you can use any of the certified like good detail models out there like night vision Excel dinov Vision Excel um yeah that's the two I can think of at the top of my mind all the links for what you need will be in the description including the lat and consistency SD XL Laura all right so let's get started so first things first obviously we're going to load our checkpoint I am apologizing for my voice by the way I'm sick um but I'm sick so my voice sounds a little bit weird but first things first is we're going to load a checkpoint as I said in this case I'm going to just use reality check Excel all right so the next thing we're going to do is we're going to load a Laura and that Laura I haven't changed the name of it so it's going to be called py to Laura Waits this is our LCM model Laura so we're just going to Simply connect the model and the clip and from the model we're going to drag out a h sampler we're going to drag out the positive and a negative clip text and codes this will be our positive will be our negative we're going to connect them to the appropriate notes oops now we're going to Dr out latent image and choose empty latent image next we're going to drag out a latent and use V vae decode wow I can't speak today oh we're going to drag the VA and connect it next is a preview image and that's it for the basic setup which is the basic image setup but here's where the LCM Laura configuration differs instead of your regular sampler names we're going to choose the LCM one and the scheduler is going to be sgm uniform steps are going to be at I prefer eight but like six I'm going to put it at six four gives you the fastest results I feel like six gives you the better results though you can go anywhere from four to eight steps or 2 to eight steps in the CFG scale we're going to put at 1.3 so an important thing to note is anything below one and Below 1.2 is not going to take into consideration your negative prompt so if you're a heavy user on negative prompts or use negative prompts in general I recommend keeping the value above 1.2 I found that the best value is CFG of 1.3 personally I find that the images come out looking great that way so let's do an example prompt all right and we C The Prompt initially it's going to take a bit cuz it needs to load the checkpoint and who we almost missed it ah there it is I forgot to add these prompts but right now I want you guys to check it out cuz this gives you a better look into how long it takes I'm going to click on C promt and you'll see the magic did you see that we can see it took less than a second it took 0.97 and that's on six steps I personally prefer eight steps for the best quality which takes around 1 second which is insane it's insane uh it's important to mention I have a 3090 however it seems that these results are still incredibly quick on Lower vams let's change the prompt a little bit all right and we're going to do a photo of a tiger would you look at that you guys that is so quick and such high quality too um let's do a photo just for the proof of concept here of a blue tiger and see what happens it gave blue eyes earlier I did a pink tiger so I'm going to try that see what it does here yeah there you go as you can see the photos are really really really good quality and incredibly quick yeah all of this under 2 seconds and this is with eight steps it's insane I hope you guys learned something new in the video I'm going to mention again that all the links that you need will be down below in the description thank you so much for watching if you like this video please just like And subscribe it helps the channel a lot I hate asking for likes and subscriptions but honestly it just helps the channel I'll see you guys on the next video bye

Channel: Neural Alchemy

Views: 2,310

Keywords: frontend, technology, svelte, sveltekit, react, reactjs, nextjs, javascript, css, html, learn frontend, javascript tutorial

Id: a6hMeXQSNTc

Length: 6min 38sec (398 seconds)

Published: Sat Nov 11 2023