Magnific AI Upscaler Free Alternatives! Krea and Comfy UI Workflows

AI upscales like Magna have been making rounds across social media and rightfully so the results from these upscales are pretty miraculous you can see these before and after images and some of the details that they enhance these images with are pretty mindblowing but these aren't actually upscaler traditional upscaling works by taking the pixel level data the actual colors and multiplying those so you can create a larger image but these upscaler are completely different what they're doing is actually running a model based on stable diffusion and they're generating details that don't exist in the original image one of the best examples I have is of this Laura Croft you can see that it takes this polygon version of Laura coft and turns it into this beautiful full resolution High Fidelity photo of her but that's not possible none of that photo none of that pixel data exists in that original image and that's where stable diffusion and AI generative art comes in and I'm going to show you three different free tools to achieve similar results first I want to show you some examples on the magnifi website take a picture of this house out by the ocean as an example and you can see as you slide the image to the left it's completely changing the flowers that are in the flower pot none of that data exists in the original image it's also adding blossoms to the flowers on the roof and adding other details that just simply don't exist in the original image now don't get me wrong this result is stunning but it isn't upscaling it's more reimagining and this is another great example watch closely as I move the slider across the bread you're going to see changes in the texture and even in the Contours of the bread itself look at this end piece there's no break in it it's just crusted but when you scroll over it you can see that it's almost as if it's been sliced so you're getting a lot more detail and this looks fantastic but it isn't that original image and I actually don't think that that's a problem there are a ton of use cases for taking a low resolution image and bumping it up if you're using it for something like b-roll or for a presentation and you just don't have a high enough resolution image so I see a ton of use cases for this technology and this is literally the only reason I don't use magn there's no free version it starts at $39 a month and that's just simply too high for my usage and I know their GPU costs for running this are through the roof so I don't blame them for this but this isn't a debate about their business model I want to show you how this is actually being done on the back end and that brings us to our first free competitor and that's Korea Korea is in open Beta so at some point they're going to have to have a business model for this they're going to have to start charging but for now you can do this completely free and they have two different things that are really cool realtime image generation so as you type it actually generates an image or upscale and enhance and today we're going to focus in on that upscale and enhance the user interface for Kaa is really simple you've got this area to upload an image place where you can put in a prompt to describe your image how much you want to upscale the style some sliders to control the AI strength and the resemblance strength and then finally a negative prompt and I happen to have the perfect image to start with it's a really really low resolution thumbnail from my 3D printing YouTube channel as you can see the image quality on this thing is awful it's lost almost all the detail you can't really tell but I'm holding a hydroponic cup from one of my 3D printers and those things coming through my hands are actually the roots of the plant for a prompt I'm going to help it out a little bit here I'm going to say a young adult man in a backyard standing on the patio holding the fine stringy roots of a tomato plant from a hydroponic garden pot upscaling Factor we're going to do 2x upscale for the style we're going to go photo realistic and then we're going to go ahead and just crank up the AI strength and the resemblance strength all the way I'm not going to put anything in the negative prompts going to click enhance and here's the result look at the details as we start to slide over this now obviously it loses the fact that this is a photo of me but this is where I'm talking about these aren't upscaler they're reimagining things here's where it gets a little wonky you can see that it replace some of the roots with another hand this is where stable diffusion sometimes gets a little bit weird but I got to say back in the back there that's actually what my hydroponic tower looks like this is a pretty good result overall and something that created a somewhat usable image out of something that you couldn't use before and I think by adding a negative prompt and refining our prompt a little bit more we could get an even better result and Korea provides their own examples as well like this chocolate cake that is completely reimagined but as you can see pretty much follows what you'd expect now there are some changes just like with Magna you can see over in here it changes what looks like maybe nuts or something over to some type of fruit similarly here's a really low resolution photo of a frog and it changes it completely keeps most of it although you can see it loses some of those blue colors from the original image it's still a really cool result but what if you want to run these on your home computer that's where some software called comfy UI comes in comfy UI is a modular node based stable diffusion system what that means is you can string together multiple different types of commands to automate an entire workflow let's say your typical workflow is to generate an image upscale it and then save it this can do all of that in one pass without you having to run multiple different commands and if you go to websites like open art. you can get other people's workflows you can even share your own and these can be really powerful everything from creating stable diffusion animations or if you watched my video on D XL turbo we also used a workflow generated in comfy UI if you're one of my patreon subscribers I've got a oneclick installer for comfy UI otherwise let me know in the comments below if you'd like me to do an entire walkthr on how to install comfy UI and get it running for the next two examples we're going to look at some workflows created by Olivio cus he's a great YouTuber and I genuinely hope that you watch his channel And subscribe to it if you haven't he's created this photo portrait upscaler which you can run in collab directly on the open AI Art website or you can download it and run it locally on your own install of comfy UI there's also portrait reimagined upscaler which is more of that magnifi style where you can take a really low resolution image and make it something completely new let's take a look at both of those now for this first example of the photo upscaler we're going to use collab so we don't have to install or run anything locally on our machine I'm going to use the same file that we used in that last example for Korea and I'm going to give it the same prompt now you can see this is a pretty complex setup it goes through all these different nodes it's upscaling 4X normal here it's running a bunch of different refiner models it's doing a lot of different image masking types using control Nets and then it's finally running it through stable diffusion with the realistic Vision 5.1 safe tensor model and here's the end result you can see it started with that same low resolution image and it ended with arguably a better result than Kaa in the fact that at least the hands are normalized and you don't have any extra body parts now it is a little bit blurrier and a little bit lower resolution but actually I wonder what would happen if we ran the resulting image through the upscaler once again let's try that go ahead and take this image and save it we'll go back to the original node choose a file to upload we'll select that new uploaded version and you'll click on Q prompt that's going to kick off the entire process once again and you see down here at the bottom here's the original image it created two two different control Nets the one on the left is a depth map and the one on the right just sort of gets the outlines and the edges of all the detail in the image and here's our final image we're getting a lot more clarity on some of the details in the plants and the leaves you can see the tree in the background has a lot more detail in the new result than it did in the old one the face the hairline the clothing everything else looks less blurry there's still some really weird stuff going on with the plants around the hydroponic tower and it looks like the eyes are starting to get a little wonky but we could always take this run it through a control net face swap in something like Focus we could probably fix all that stuff too pretty cool for the second one the portrait reimagined upscaler there's no option to run this in collab directly on the site so we're going to download the Json file we're going to jump over to my own comfy UI install running on my Local Host machine we're just going to load the configuration file go into the manager and install missing custom nodes once that's done we'll upload a file in this case we'll use the one generated from the last image set and we'll use the exact same prompt that we used on our last one click on Q prompt and that's going to kick everything off and get it running all right here's the result that it came back with now at this point it's realistically just an entirely New Image you can see that the roots have been replaced with actual tomatoes I'm holding a tomato that's been cut in half and then there are tomato pots everywhere it's replaced The Back Fence with a garden it's replaced the concrete wall with what a appears to be some sort of SN and wood columns but it's still pretty cool this is almost like a control net that really just uses the original image shape and depth map is guidance and then completely reimagines it so reimagining upscaler is kind of a perfect name for this thing and these are really just two of hundreds of different workflows that you can run through comfy UI hopefully you learned something today all the links to everything we tested today are going to be down in the description below let me know what you think in the comments and if you want me to do a deep dive on any of these subjects even a comy UI andall guide otherwise I'm Brian love it this is all your Tech AI thank you so much we'll see you next time the in the break down the crown basic to complex never let you down all your tax a i earning the renown

Channel: All Your Tech AI

Views: 30,643

Keywords: magnific ai, krea ai, ai art, magnific ai alternative, free ai tools, ai image upscaler, magnific ai image upscaler, magnific ai upscaling, stable diffusion, generative ai, increase photo size, low quality photo to hd photo, how to depixelate images in photoshop, magnific upscaling, ai upscaling video, ai upscaler, best ai tools

Id: uPQE0IlQZd0

Length: 10min 21sec (621 seconds)

Published: Wed Jan 31 2024