RIP Midjourney? Open Source & FREE AI Art is About to Take Over! DIRECT COMPARISON

let's be honest viewers taking a step back and looking at the larger picture mid Journey has somewhat of a domination over the AI image generation space sure there are smaller communities in the AI world that still use things like stable diffusion because of its open source nature but the original stable diffusion can't come close to the Fidelity that mid-journey produces however what we're looking at today will give mid-journey a serious run for its money not just in the Fidelity department but with a few extra exciting features that mid-journey just can't hope to add as well as the fact that the AI we're looking at today well it's not going to cost you any money this might look boring but this is actually really exciting stability AI launched sdxl 0.9 and this is a serious Leap Forward in AI image generation for a number of reasons you might be thinking back wait sdxl was already released wasn't it well actually it was just in beta in April this follows that up and actually produces a massively improved image and composition detail over its predecessor in beta it's already available for you guys to use in a number of places and we'll talk about that in a little bit but the research weights are now available and an open release is coming mid-july and that will move into sdxl 1.0 it's got the classic stable diffusion stuff we love the ability to run on a modern consumer GPU yes that's right you can run this thing at home completely free and of course that open source release God people love the open source capabilities right mid Journey wouldn't dare let you access all of their Secret models but stability AI will give you the whole can of beans so sdxl has the ability to generate hyper realistic Creations for films television and music as well as offering advancement for design and Industrial uses all of this places sdxl at the Forefront of real world applications for AI imagery which is basically what we've seen with mid-journey so far and viewers that's the real question we want to answer today right if sdxl is the open source Champion that it claims to be can it Dethrone the reigning proprietary Powerhouse mid-journey well we're gonna find out so we're gonna get into some quick examples this highlights the exponential leap that is sdxl 0.9 over its predecessor just the sdxl beta so as you can see on the left here this is the original sdxl beta not really that impressive a lot of people consider this better than the original stable diffusion but then again the original stable diffusion had a lot of we can call them mods so these mods would be things like control net or dream booth and they would increase the Fidelity in the specific use cases of what you're trying to create right out of the model there is a massive difference with the sdxl 0.9 model I mean that is something I would honestly expect out of mid Journey pretty much I mean maybe I'd expect a little bit more detail out of my journey but it's very coherent look at this alien's face there's not a lot of artifacting going on the background blur looks realistic the lighting looks realistic reflectiveness of his clothing is absolutely ridiculous it's a little bit convoluted and weird in certain areas but that's something that we still see with mid-journey these models are not perfect yet and the prompt is pretty simple here it is simply aliens walking Among Us in Las Vegas scratchy found film photograph moving on to the next example here we've got a serious example of the accomplishments they've made remember this all runs on a consumer graphics card mid-journey I can almost guarantee does not whatever models mid-journey is running take quite a long time to generate and I don't think they'd run on just a few gigabytes of vram and again this looks like something I would expect out of mid-journey and any rate to The Prompt here was a wolf in Yosemite National Park Chile nature documentary film photography I mean we could see the original sdxl beta again this is something that might be of better than the original stable diffusion but the eyes are screwed up the detail is lackluster the the only good thing about this original image is the framing this one is almost Flawless almost looks like a completely real photograph maybe a little bit weirdness down here other than that you know the wolf's face is symmetrical it looks real the lighting is dramatic it looks like it's from a nature film photography thing it's almost perfect mid-journey level I would say but these are cherry-picked again we're going to be doing some raw testing this one shows a little bit of the accomplishments in how natural the photographs look rather than just trying to accomplish everything The Prompt asks for it's trying to blend it in a little bit more realistically so the prompt here was manicured hand holding a takeout coffee pastel chili downbeach Instagram film photography and some of these by the way guys have negative prompts as well which we don't really get with mid Journey do we the negative prompts essentially allow us to negate things from our final images so if there's something we really don't want in there we can say please none of that anyways in this image here it's quite evident that it's really just barely managing to capture everything in the final image the hand is definitely there all and the hand is pretty screwed up too on the fingers over here but it's just kind of holding it out like yep the hand is there and it's holding the coffee that's about all that's happening the beach is definitely looking pretty nice though and yeah the coffee looks alright but that no one holds a coffee like that and that definitely isn't an Instagram post that's that would be a very weird Instagram post this one honestly I can't even tell that it's not real it looks like something I'd see on Instagram really nice beautiful beach lots of details too with the trees over in the corner and the waves and everything really phenomenal honestly and the hand is definitely more manicured and this image in comparison to this one and the skin I gotta say the skin details on this is pretty phenomenal that is quite shocking welcome viewers to the mid-journey Community Showcase this is where you can see all of those wonderful mid to Journey Generations the trending ones the hottest and yeah we can see mid-journey is phenomenal this is a serious competitor there are actually some things that I personally don't like about mid-journey such as the fact it's honestly not too fantastic at following your prompts it really does seem to highlight overall aesthetic pleasure over actually obtaining every single thing you ask for in your image however the aesthetic pleasure is definitely phenomenal I mean all of these images are quite something to look at let's be honest the big problem here with mid journey is the fact that it's not open source there are no little modifications you can add on to it you can't run it at home on your own graphics card you can't really train it on your own face such as you could with dream booth and stable diffusion and guess what sdxl is going to have all of those capabilities right out of the box without any modifications though how does sdxl 0.9 handle against mid-journey currently this is where you can try stable diffusion XL 0.9 this is the website known as clip drop by stability AI yes it's completely free to test out and use as many prompts as you want there's also the style setting for your prompts here with stable diffusion XL mid-journey doesn't really have specific Styles like this so I'm going to select no style for these examples this is the mid-journey image that I want to compare to First all of these images are selected at random but yes it's a very nice mid-journey generation a woman age 35 full body silver wavy hair wintry eyes casual leggings in a T action pose Dynamic walking pretty cool stuff going to be difficult to compete against let's go ahead and generate with clip drop I'll just go ahead and select our favorite one which which I think is this one it does some automatic upscaling to make the image 1024 by 1024 the same resolution as mid journey and here's our final result comparing the two images side by side I gotta say they both look pretty good interestingly enough stability ai's sdxl model decided to make the woman a lot older I would say that she is definitely older looking than 35 but this woman over here looks a lot younger than 35. either way yes they are both wearing the Casual leggings and look like they're jogging the hair is Flowing the eyes are the same color even of that with that wintry color hair is the same color and there's a bunch of cinematic stuff going on as well and I gotta say both images look very much cinematic uh there's definitely some weirdness going on with the shirt though on the sdxl 0.9 side but I am very very impressed about the amount of detail that is coming through on the sdxl 0. 0.9 side I mean truly a feat in comparison to the old model this is the best image that I could generate with the older model and uh yeah that is a lot worse so this is a massive improvement over previous stuff and this is all going to be available for you to run on your GPU at home for completely free or even use on other websites for completely free just like stable diffusion where mid-journey is going to cost you money and of course don't forget about all those extra modifications and additions that you can add on to these stable diffusion models so you know the images are close enough where are you really willing to pay for that tiny amount of extra quality you get out of mid-journey I don't know all right here is our next image comparison this was two geese standing next to protecting Knights cute Goose princess Goose princess Goose goose night night Paladin anyways it's just a bunch of goose night words they want two night geese essentially so again the comparison here they're both very very good I would say these are both what I would describe as geese Knights the mid Journey geese definitely have a higher Fidelity to them I don't think there's any denying that there's a little bit more color it's a little bit more aesthetically pleasing there's more detail going on in here I I will not deny but I am still very impressed by the coherency of these geese Knights still there is a lot of detail going on here we don't see this much detail from many other models in the space in fact I would say this is like the second best right in comparison to the amount of detail you get here and again this one's going to be free for you guys and has all those open source capabilities a little bit weirdness going on with the legs here I will note but still there's some hand weirdness too going on with mid-journey I move my head you'll be able to see that her hands are a little bit weird either way both are impressive the mid Journey one is definitely a little bit better so for this one I decided to go with an extremely realistic mid-journey Generation The Prompt here was kids in a classroom wearing Apple Vision Pro glasses reporter Dodge style realistic photography both images honestly look very very much like photos are very realistic a little bit of weirdness going on with the eyeballs on the sdxl 0.9 side but I gotta say the hair looks really realistic and the clothing as well it's funny because they're like apple Vision Pro glasses and it literally put Apple's literal apples inside of the image where you don't get that on mid-journey Mid Journey definitely fits the prompts a lot better for this one and it somehow understood what the uh the realistic VR glasses are kind of supposed to look like and yeah the overall image looks a little bit more like a true to life photo still you know we did kind of get retro futuristic glasses on all of these uh characters on the stable diffusion side so that's pretty impressive again I'm impressed with the open source side it's a little bit behind but dangerously close to mid-journey beforehand it used to just be a blowout mid-journey would win every single time even though these models were open source now you know things are starting to get really close the gap is is coming together and we haven't even tried stuff like the different styles or even things like control net for example those little mods that you can add on to stable diffusion models so there's an interesting comparison with these two images right here the prompt was female model incredible future close-up chatoyant eyes honestly I think that both images got the prompt correct because the prompt isn't very specific both of these could be considered close up imagery obviously the mid Journey one is a lot more close up and decided to do something entirely different and this shows the difference between the models creatively and how they want to express the prompt obviously we still got a really really phenomenal mid-journey image here I mean look at the detail but still this sdxl 0.9 image is really really quite good the eyes are still popping out and it's still very close-up image and honestly it captured the future aspect of the prompt quite a bit better then mid-journey did so this is you know sometimes it just comes down to personal preference which one do you like better the prompts here was creative character 3D render anthropomorphic lemon character relaxing on the beach Pixar and I gotta say I like both of these results a lot it's fair to say though that mid-journey significantly won out with this one I mean look at how this lemon character is kicking his feet up he's got even five fingers he's really enjoying himself he's sipping a drink definitely on the beach lots and lots of detail you can really see that lemon skin and over here on this side we still got a pretty fantastic lemon character although his hands are a little bit more screwed up and he's just overall not as fun and creative as this side but again it's important to Remember When comparing these models every single AI model needs to be prompted a little bit different so I think that overall these models trade blows mid Journey definitely has an advantage when it comes to detail because it's generating in a natural 1024 by 1024 resolution where this one upscales from 512x512 still though it is hard to beat free and open source I really do think that you can get images out of stable diffusion XL 0.9 that are truly at that mid-journey level I mean take a look at that we definitely could get something like that out of mid-journey no doubt we can even go ahead in here and change this style if we want let's try the fantasy art style and regenerate the image these are different features and benefits you just don't see with mid-journey check that out the fantasy art style that did a number on the image it made it look really really cool check that out completely transforms The Prompt clipdrop is a lot of fun to play around with viewers I'll definitely link this down below so you can test out stable diffusion 0.9 again we're expecting that full release next month for now though you just got to use it on the site the key driver for this advancement in composition for sdxl is the significant increase in parameter count this is essentially the sum of all weights and biases in the neural network itself that the model is trained on sdxl 0.9 has one of the largest parameter counts of any open source image model boasting 3.5 billion parameters for the base model and a 6.6 billion parameter model and symbol pipeline which is the final output that is created by running on two models and aggregating the results second stage model of the pipeline is used to add the finer details in the generated output here's the kicker with the system requirements you can run it on a modern consumer GPU you only need Windows 10 or 11 or Linux 16 gigabytes of RAM and this is the ridiculous part the graphics card which again does the main processing of these images you need a GeForce RTX 20 series card or better equipped with a minimum of 8 gigabytes of vram that's it eight gigabytes viewers that is not high-end stuff at this point eight gigabytes of vram is expected in fact it's necessary these days for a lot of modern games so a lot of you viewers at home are are going to be able to run this on your own machines like a lot you viewers at home if you want to figure out if you can run it at home right now hold down your Windows key press the x button a little menu such as this will pop up go ahead and press the task manager go over to the performance Tab and go down to a GPU and then look for dedicated GPU memory as you can see I have an RTX 4080 so I have 16 gigabytes of GPU memory but we only need half of that to run this model and honestly viewers I just upgraded to this RTX 4080 which Nvidia graciously sent me so I used to have eight gigabytes of vram this is the same exam act vram requirements as we saw with the original stable diffusion and also if you are a Linux user you can use a compatible AMD card with 16 gigabytes of vram and typically newer AMD cards actually have more vram than Nvidia ones so if you're on team red that's actually a pretty good sign so the availability here like I said and we saw earlier it's available on clip drop for completely free which is really fun you'll be able to access the stability AI API again to build different things like let's say you wanted a video game that generates images or any other kind of application that generates directly with the API dream Studio customers will also be able to access it and other leading AI image generation tools like night Cafe for example right now it's in that research purpose before the final release and the code will be available on GitHub with the open release later this month viewers I know it you know it AI open source technology needs to be the leading type of AI Tech time again we've seen honestly that a lot of companies are not able to facilitate what the public truly wants some AI technology is very very dangerous this definitely can be one of them but I would say the image Generation stuff isn't too bad right now and open source allows for again the community to really do whatever they want with it it's not controlled and locked down by a central defining force a central company we can modify it we can edit it we can make it better we can make it do very specific things we can make it cheap we can make it accessible for everybody the technology isn't limited or held back by nearly as much as it would be otherwise the barrier for entries are lower that's why I love to see that an open source AI image generator is really taking the fight to mid-journey a special on that Aesthetics front which really seems to be important for a lot of users right now with the popularity of mid-journey I'm excited for this let me know what you viewers think down in the comments below and I will see you in the next one check out the disc server check out some of my other videos and goodbye

Channel: MattVidPro AI

Views: 197,944

Keywords: Ai Art, ai image generation, midjourney, stability ai, open source ai art generator, open source ai, mattvidpro

Id: 9VUBlCv6Z8g

Length: 18min 38sec (1118 seconds)

Published: Tue Jun 27 2023