Better Than Midjourney: Openjourney Stable Diffusion

today I'm going to show you something way cooler than mid-journey and I know that's a bold statement if you're not familiar with mid-journey what it is is an artificial intelligence stable diffusion website that allows you to connect to Discord and generate some amazing pieces of artwork like this some of this stuff is just absolutely stunning but one of the primary problems with this is it is not free you actually have to pay a monthly subscription for this and on top of that you have limited number of images that you can create based on the subscription plan you can see that based on the plans it goes anywhere from free trial with very limited resources all the way up to a 60 a month Pro Plan which offers 30 hours a month of fast GPU time and 12 concurrent jobs but I can tell you just from the past couple weeks of playing around with stable diffusion I probably spend that much time just in a single day so let's find a better way and jump right into it so what I'm going to show you here today is how to run mid-journey actually as a low local service on your own PC so you can generate unlimited free images and this is based on training data off of actual mid-journey images so the results should be very similar to what you get with mid-journey the main difference being of course that you don't have to pay for it now in order for this to work you need to have already followed my tutorial on setting up automatic 1111 on your local Windows machine if you haven't done that be sure to check it out after this video so that you can follow along and with that out of the way we're going to take you over to hugging face hugging face is an amazing website that has a whole bunch of different training data sets models and everything else you can use for stable diffusion the best part is I found this open Journey Open journey is actually an open source mid-journey trained model that you can take and download and drop directly into your local hosted version of stable diffusion using automatic 1111. it's super simple to do all you have to do is go over to hugging face I'll draw drop a URL down below and you can search for open Journey you can see in the models section there are a number of different models that come up so you can try different ones out this happens to be the one from prompt hero that I've been using for the past couple days simply download the checkpoint file for this and then go over into your models directory inside of your stable diffusion install and simply drop this in there it couldn't be any simpler than that the other cool thing about this website is you can search for other so we'll say mid journey and you can see there are a whole bunch of different models that you can train for them so if you go to the model section we'll just go ahead and click enter here you can see there are 21 different data sets just for mid-journey alone pretty awesome go ahead and test some out I've got a Discord Channel set up now so hit up that link and let me know some of your results maybe you found a model that's even better than the one I'm testing so once you've got that checkpoint file installed and you've restarted your automatic 1111 UI let's jump in and see what some of the results of this new model are and to do that I'm gonna head over to a couple of prompt sites that I found that have some mid-journey style prompts the first one we're going to check out is creative indie this site's pretty cool because they actually have a list of 644 mid-journey prompts so most of these you'll be able to drop right into automatic 1111 and get some pretty similar data out of it so you can see a few of these examples here and they have the prompt and then the results over here let's see what I came up with from a couple of my own test prompts up in this stable diffusion checkpoint in the upper left hand corner you're going to see this drop down box that's where you select the model that you're using for automatic 1111 in this case mid-journey V4 checkpoint is selected so I am using the mid Journey model for this my prompt is a surreal Underwater World in the style of HR Geiger I don't have a negative prompt in here my sampling method I went with DPM plus plus 2m Keras sampling steps so this is a number of times that the AI model the neural network iterates over the image to improve the quality so I did crank this all the way up to 150 other just wanted to see how much detail we could get out of this and then I did the high res fix and I used an esrgan 4X upscaler this just increases the resolution of the image from 512 by 512 which is kind of the default 2024 by 1024. batch size went with one and batch count 16 so that we have 16 images in total that are generated so let's go ahead and take a look at the results you can see these came back they're really cool looking it's almost like a pencil line art the highly intricate details of all these images and I think it is it's really fascinating the way that it came back with this style that one's pretty cool and so this was the first example prompt that I went through I'm already kind of excited by this point right you're starting to see that okay this could generate some pretty cool stuff so I went from that to this scene so this is a prompt a historical epic battle scene in the style of Frank franzetta and you'll notice a pattern here oftentimes what you'll end up doing is you select an artist in the prompt and that'll start to kind of stylize the generated images in the style of that artwork so you can get back some really specific detailed images that way so take a couple looks at these and you can already see these are definitely epic battle scenes holy cow the the intricate detail on these again kind of blows you away wow some of these are pretty insane looking I can't even believe that and for this one really the only changes I made again I said use the same upscaler I use the same high-res fix ranked up the sampling steps to 150 same sampling method everything else nothing really changed there really cool result and then the third one I did from that website was this a steampunk airship in the style of Jules Verne I must have been going for something where I just wanted those really highly intricate details because as you can clearly see every single one of these images was kind of heading in that direction very cool stuff now here's another awesome tip there's a bunch of websites that have various different methods of generating prompts they help you kind of Step through the prompt building process because that's honestly how you get the most out of stable diffusion is by having a good prompt one of those websites is prompt hero prompt hero is really cool because they actually have this entire section on the best open Journey prompts and that's what we're using for this this is open Journey you can see up here in the menu they have mid-journey chat GPT stable diffusion so you can go through and you can start scrolling through some of these really cool images that are generated and then you can pull one of these up and say this one for example and it'll actually tell you the prompt it'll give you the resolution the seed the number of steps even the sampling method that was used and you can just take this information and you can plug it directly into automatic 1111 and you should be able to get back a similar result to this so I went ahead and did that with a couple of prompts I found on their website this one I thought was really cool is highly detailed robot and so we dropped that in you can see the full prompt here full body cyborg full length portrait detailed face symmetric steampunk cyberpunk cyborg intricate detail you get the point right it's very intricate and highly detailed this one does have a negative prompt so you want to get rid of things like Blurry fuzzy bad Anatomy disfigured cranked the sampling steps up to 150. this time I did not do the high-res fix but I did do a slightly larger height I did 768 pixels and this is some of the stuff that it came back with and you can see that it's overall very similar to the style that was shown up on the website so this is cool because if you're looking for a specific style you can find a website that has some prompts and using that you can get to a place where you have a pretty good representation of what you are going after with a little bit less effort and then from here you can go back and you can just start to refine The Prompt and you can refine the details of some of these images until you get exactly what you're after I thought that was a cool one next up I went with this prompt I thought this was uh just a stunning image it almost had a photorealistic photography feel to it so you can see here the prompt mid-jorny mid Journey V4 style medium length white hair young South Korean girl so this whole long prompt and then we've also got our negative prompt over here someone saying didn't really change any other settings other than the height and width so let's see what this came back with and you can see it's sort of mixed results right so this one was just a little off to me I don't know about the Wisp of hair right down the middle of her face but you know that's kind of interesting obviously this one has multiple faces and I wonder if the face fix feature would actually get rid of that unsure again another one where there seem to be uh doubling up of the bodies or body parts but I did think this one was actually pretty good out of the results side it was a very highly detailed picture and I thought it came back very nice so you have it it's basically like having a free version of mid-journey right there on your home Windows PC hopefully that helps some of you out hit like And subscribe if you'd like to see more of this type of content join me in the Discord let me know what you've been generating I want to see some of your images we'll talk soon thanks so much

Channel: All Your Tech AI

Views: 183,612

Keywords: stable diffusion, ai art, stable diffusion ai, midjourney ai, midjourney ai tutorial, midjourney prompts

Id: gLk_DCmLFYM

Length: 9min 52sec (592 seconds)

Published: Mon Mar 06 2023