MidJourney In Trouble? Open-Source and FREE AI Generator SDXL Is Stunning (Comparison)

if you want to create absolutely stunning AI generated images that are similar in quality to mid-journey absolutely free and nearly unlimited this video is for you so strap in we're going to take a look at the new stable diffusion which is absolutely stunning I'm going to tell you about it then I'm going to show you how to use it and then we're going to do some direct comparisons let's go this is the blog post today stability AI announces sdxl stable diffusion XL 0.9 a huge leap forward in AI image generation so stable diffusion XL beta was released in April and just a few months later stable diffusion XL 0.9 produces massively improved image and composition detail over its predecessor the rate of improvement of these open source models whether you're talking about large language models for text or AI generative art models these are absolutely huge gains every single day seemingly and the best part about it it's open source it's completely free 3 and soon you'll be able to run it on your local computer despite its ability to be run on a modern consumer GPU sdxl 0.9 presents a leap in Creative use cases for generative AI imagery first let's take a look at some examples from the previous beta to this new version on the left we're seeing the beta on the right we're seeing the new version these are the same exact prompts and as you could tell the left it's pretty good and the right is so much more detailed so much more color there's bokeh in the background it just looks so much better let's take a look at this second image the left again is the beta version and the right is the most recent release a wolf in Yosemite National Park Chile nature documentary film photography so this one looks pretty good although the wolf is far away and the details are not great the log doesn't look super real but if we look on the right this one is fantastic it's gorgeous you can see all the little hairs the details are phenomenal the eyes look perfect this is a really great image here's another one aesthetic manicured hand holding up a takeout coffee pastel chili Dawn Beach Instagram film photography now to be honest I think this left one looks really good the only criticism I have is that you can tell that there are lots of fingers here many more than what humans have ai generative art has always struggled with hands mid Journey's new version has really solved it but it looks like now stable diffusion XL 0.9 has solved it as well this hand looks Flawless and the sdxl series offers a huge range of functionalities that extend beyond basic text prompting image to image prompting in painting which is basically taking portions of the image and replacing it with generative art and out painting which is constructing a seamless extension of an existing image basically taking a little piece of the image and creating AI art extensions of what's around it so how did they do it the key driver of this advancement in composition for sdxl 0.9 is its significant increase in parameter count over the beta version as sdxl 0.9 has one of the largest parameter counts of any open source image model boasting 3.5 billion parameter base model and 6.6 billion parameter Ensemble Pipeline and the beta version for comparison has 3.1 billion parameters and uses just a single model sdxl uses two models clip models they're called including one of the largest open clip models trained to date which beefs up the processing power and you get upgraded realistic imagery greater depth and resolution up to 1024 by 1024 they're going to be releasing a blog post talking about all of their advancements soon here's another image absolutely stunning galaxies in the background a little Galaxy in this bottle a really beautiful image now here are the system requirements despite its powerful output and advanced model architecture sdxl 0.9 is able to run on a modern consumer GPU you need Windows 10 or 11 or a Linux operating system 16 gigabytes of RAM now that that's not vram that's just regular Ram which 16 gigabytes most modern computers have you need an Nvidia GeForce RTX 20 graphics card which is kind of mid-range lower mid-range graphics card and eight gigabytes of vram so very doable for many video cards and really a minimum requirement for any modern video game right now you can use stable diffusion XL on clip drop which is what we're going to be testing it on today and the API in dream Studio customers can access it as of three days ago so they'll be able to use it as well sdxl 0.9 will be provided for research purposes only during a limited period to collect feedback and fully refine the model before its General open release the code is available and you can already use the code today so they've released the code it's on the stability AI GitHub page you can download it you can play around with it right now I'm not going to do that but maybe I'll create another video showing you how to get it up and running on your local computer and what's next stability XL 0.9 will be followed by the full open release of sdxl 1.0 targeted for mid-july so very very soon now here's the stable diffusion XL model on clip drops website let's take a look at some of these examples I mean these are absolutely gorgeous we have anime style right there realistic with a skeleton there hyper realism right there tilt shift effect unbelievable these images seem absolutely on par with mid-journey first let's take a look at pricing as I mentioned it is completely free mid-journey you can't even test for free and you get 400 images per day so it's basically unlimited I don't know about you but I've never even come close to generating 400 images in a single day and with the free version you get a bunch of other AI features background removal cleanup pictures relight which we'll take a look at image upscaler really cool now let's do some direct comparisons I'm on the mid Journey website I'm gonna grab some of these prompts and then I'm going to test it out and compare the images directly with clipdrop so first I'm going to test this one out pill head surreal surrealism abstract Harry Clark this one looks really really cool now I don't expect the images to look very similar because they're trained on completely separate models what I'm curious about is is the quality equal and so let's see all right there we go now as I mentioned the images really don't look like what I found on mid-journey but these are gorgeous these are very artistic as well I'd say even more artistic than what was on mid-journey I'm going to show both of these on the screen right now so you can see a direct comparison next I really like this fingerprint right here so let's try to recreate it using sdxl a large fingerprint stamp on paperboard that's a super Broad and generic prompt so I suspect we're going to get something very different but let's see all right there it is so it's black and white but I think these look really good as well let's see what happens if I try to use blue and red in the prompt to get something more akin to what we found in mid-journey okay here we go these all look fantastic I'd say this one looks really close to the mid Journey version I'll put these both up on the screen now so you can see a couple versions versus mid-journey alright last I love this image of a lion and I think sdxl is going to do a fantastic job on this one so let's test it out all right take a look at this I'd say this is okay this isn't quite what I was asking for it's much more artistic I was looking for something much more realistic and it didn't seem to be able to achieve that but that's okay all right these are absolutely stunning this is exactly what I wanted hyper detailed super realistic let's click into one of them look at that you can see all the skin textures you can see each hair and the eyebrows and the beard on the head so this is so so good now I want to show you a couple other features if you click on this top right the three dots right here you have a remove background feature cleanup imperfections relight enhanced upscale imagine an uncrop I want to test out relight and what relight allows you to do is exactly what it sounds like you get to position lights in any direction on this face so you can already tell the face is much different so if I click over to the red light I can easily increase the red light and in real time you can see the lighting on his face change the Shadows change everything so cool you can change the intensity you can change the radius and then you can change the distance and so look at this if I change the distance you can actually see the light passing over his face and changing the shadows in real time so if I turn the red light off completely and then I turn the blue light up you can see that it changes the light and the green one you can add more lights to it look at that so I'll move it around his head and in real time the whole face is updated and then you can turn the lights off and then download that image so cool so I am blown away by these images the progress of this model is substantial and just knowing that the 1.0 version is coming soon is so exciting this is free this is open source soon there's going to be no limit because you can run it on your own computer but for now you'll have to deal with the 400 images per day which is a ton mid Journey really has their work cut out for them I'm so excited to see open source AI continue to proliferate continue to get better at such a rapid clip let me know what you think in the comments if you like this video please consider giving me a like And subscribe and I'll see you in the next one

Channel: Matthew Berman

Views: 147,808

Keywords: midjourney, stable diffsion, sdxl, sdxl 0.9, sdxl 0.9 download, midjourney ai, midjourney tutorial, midjourney tips

Id: aYZyJb7mH9g

Length: 9min 33sec (573 seconds)

Published: Fri Jun 30 2023