MidJourney goes bye bye! And meet its FREE replacement!

hi everybody I'm sigv welcome to the AI Art [Music] Channel so before getting started I just wanted to say that this is my first time ever doing a YouTube video uh so I hope I won't mess it up too much so be forgiving be patient with me and hopefully uh it will go smooth and clear so without further Ado let's go today we're going to talk about a model called pix art Alpha it's a text generation model it generates amazing images high quality a lot of details great colors great composition and it can actually compete with the level and quality of mid Journey images uh best of all it's free and you can use it already today if you have a strong GPU you can even install it locally but I created a Google collab for you guys to enjoy in a matter of few minutes you can already play with the model and generate as many images as you like let's take a look so let's get started we're going to pixart alpha. kb. of course the links will be in the description for every website that I'm showing you here and going over the page quickly you can see some example photos here there are some more example photos on the bottom of the page but you can already see the quality of the images is amazing the details the colors the composition everything is very nice two strengths to this model that are a big Advantage One is the training efficiency another great uh Advantage this model has is that it is very coherent to The Prompt it understands much better the prompt uh than other models with the exception of do I think the doly 3 is is the the leader when it comes to coherency it's really hard to compete with them because they have the the Chad GPT engine behind them but looking at this model compared to the other open- source models in the free models there is no comparison and later on we will see that even the the generated images come close to the quality and sometimes even better than meid Journey H we can see here it it supports control net that allows you to use reference images uh for your Generations it controls dream Booth meaning that you can train the model to have your own objects and uh it also allows usage of Lura and uh all other features that we can use with the stable diffusion for example but also here you can see more samples of the quality of the images there is no question about it the the quality is amazing and all that's left is is just to play with it so there are a few ways you can start using this model on the website itself they give you a hugging face demo the hugging face demo works well uh but it takes a lot of time until the model generates the generation but we can an apple on a wood table and once I click run it will take some time you can see I'm on a queue nine out of nine eight out of eight it takes around 15 to 20 seconds to generate one image so when there is a big queue it can take some time until you get to your Generation Um so I created a Google collab let's we let the model run in the meantime I created a Google collab that is based on the collab that they give you here H there is a collab demo so I've taken this demo and I made it uh much more easy to use and also added some features that does not uh that are not included in in the collab they provide um I added something similar to what you find on the hugging face example uh where you can actually play uh with the settings if you'll open the advanced options you can add negative prompts you can choose the image Styles the width the height the number of steps and so on uh so we will just wait for this demo to show you the image and then we will go on to the collab and as you can see we have an amazing image great details if we open the image in a new tuab you will see that the size of the image is impressive it's 1024 by 1024 you can see the droplets on the Apple itself you can see the texture of the wood is amazing the contrast between the colors the saturation is very nice you can see the Blurred background that looks a very nice composition and what's left to do is to check out the Google collab and see how we continue from there okay so we open the Google collab uh once again the link will be in the description and as you can see it's a very clean and short Google collab what it will give you is the setup initial setup and then you have some kind of a form that you can play here and generate your images one thing to make sure before you run this Google collab make sure when you have the connection on the right side here on the on the top make sure that on run type Click Change run type and make sure that you are on GPU if you use CPU it will either not work or it will take forever to generate an image so T4 GPU this is of course a free account uh I'm not using a paid account I I don't see the advantage here here because I can generate many images before I get a timeout and if I get a Timeout on the worst case I'm creating a new account so let's go over the things you have two buttons to click here the first button you just click it and what it will do in the background it will just set up everything it needs in order for you to use the pipeline of this model once you click it it will take a few minutes around five minutes to set up everything it needs disregard all the crazy lines here if you're not familiar with the python coding uh it will uh mean absolute nothing for you and you don't really need to look at it if it bothers you you can just click here on the X so it will clean it up while running you will see that it starts downloading files it's actually downloading all the files required to run this model and what you need to take a look at is here you can see these two files that are not 9 GB each are the actual model files so this model is at around 20 GB okay so once it's done you can see a green check mark on this section here everything is downloaded and it actually means that the model is ready for work you don't have to run it every time you generate an image of course it's only once and then you can work with the model I love clicking here on the on the X below just to clean the screen so it looks clean and let's see if we can start using the um model let's actually go with the same prompt that we created before uh an apple on a wooden table and click play once again it will take a few seconds to run the model what I usually do is I start with lower steps and if I see and I like the image that was generated I will show you an example let's say I like the image that was generated I will simply take the number of seed you can see here it shows you the seed it used for generating the image and here instead of using zero if on the seed number it says zero it means it will randomly generate a new seed for each run I will put this seed and now I can raise up my steps to be something like 20 which will give me a much more um detailed information it's not a good idea to go nuts with the steps when you when you use too many steps the the image starts to be uh twisted and and the colors become too saturated and the edges becomes very crooked so you can play with it play with a number of steps uh usually when I start with small number of steps and it gives a great result as you can see I added more steps so we generated a bit of uh uh duplicated steam here so you can play with it and now for the for the battle of the modules let's go on and um compare some mid Journeys um prompts to this model so in order to do the comparison I'm going to use a site called prompt hero uh also I'll put the link in the description prompt hero is a great site it allows you to choose a model and based on this model uh search for images once you see an image it will give you the exact details of the prompts that was used to generate this image so we'll go over and look for um a nice image that we can compare let's search for example this looks like a nice image that we can compare once I click the image I can copy here the entire um prompt and we will go to our model and just put it here of course the tags used for Mid Journey are relevant so I'm going to delete them and what I love doing is make sure that the dimensions of the image that I'm generating are similar so here it's a square also here we will do a square and let's give it a go of course we will prove the C on zero so it will generate a random image we give it a run and just for the comparison look at this image and we will look at what the other model is generating consider the fact that this is Cherry Picked meaning that the person who created this generation probably ran it a few times until he got the image that he wanted and liked the best and here we're going to look on the first image now I know you're wondering why the hell it became a skeleton but if you look at the prompt actually there is the word skeletal and skeletal might hint the model to create a skeleton but we can look at the other details of the model you can see the amazing details of of the clothing and the skeleton themselves are very high details the colors are amazing the composition is outstanding I can easily see that it's an a great way uh a great competition to the mid Journey results this is the mid Journey results let's give it another run and see if it will generate the clowns without without the skeleton faces and on the worst case we can just remove the word skeletal and it could probably be the same let's give it a go yeah and again it's got something with the skeletal so let's just remove the word skeletal and run it again this is actually part of the model that is more coherent uh than mid Journey mid Journey completely disregarded the skeletal tag and the pixart alpha actually is a more coherent image generator than M Journey so we see once we took away the word scalet tail we get a similar image amazing faces amazing makeup the colors are outstanding and now I will show you some more images and you'll be the judge whether or not you like mid Journey better or pix art and of course remember that pixart is free Okay so let's see some more images mid Journey versus pixart Alpha uh consider the fact that mid journey I took all the prompts from The Prompt hero website that we saw before it's all cherry pecked and it's probably been several iterations until the image was uh selected and here I just ran a batch of four images and took the best one out of the four nevertheless the results will speak for themselves I think and we'll start with an image of an alien full body terrifying Fierce and threatening highly detailed and realistic this is the mid Journey image which without a doubt is a beautiful image it's some kind of a an alien with an armor suit or or whatever it is and this is the Pix art image if you ask me this image is much more fearful than uh the the previous alien it's much more uh um threatening and it looks super realistic look at the details of the veins and and the tube there's some kind of a tube going into his brain uh it kind of reminds me the alien from the movie series aliens it's it's an old uh movie series uh but it's without a doubt a beautiful generation and let's go to the next image so the next image is a Gothic heavy metal guy rocking out shattering the skies above sketching the style of gritty Urban realism gath course sketch Peg so this is the Min Journey image without a doubt a beautiful image look at all the details in the background it gives you a sense of a shattering Sky indeed beautiful image look at the hair of the dude and this is the image from um pixart I think that M Journey did a bit of better work here uh and added a lot more Det details but without a doubt this is a sketch the details here are awesome and it pretty much gives you the the sense and the feel of of the prompt and amazing the next prompt is an enormous snow monster standing on the ice in the style of an Epic Fantasy scene montore explosive wildfire teren and many many many many words that describe the and you can see that the mid Journey image is amazing the details of the fur of the of the monster you cannot confuse and it is clearly a snow monster uh it looks beautiful look at the the texture of the teeth really M Journey did a great job here and this is the image from pixart I also think it's a amazing image look at the details of the mouth it looks as very realistic it a bit gives a sense of a a 4K game game character uh but without a doubt it looks amazing look at the scales and the the the fur the ice cold fur of the creature uh if you look at the hands in the mean Journey image uh it did a a bit of a duplication in the fingernails or the claws of the creature and and on the Pix art actually the the physiology of the creature is better uh both images are amazing um you'll be the judge the next image is a hooded and thic man wolf in a metric style Digital World surrounded by flowing binary code add blue glitch effect from striking surreal look blend in geometric patterns for complexity metrix digital rain this is the mid Journey image it looks amazing it looks like a poster you can hang on your wall I really like the the pixellation it did on the top of the hood it's it literally gives a sense of of computerized character of a digital character and this is the pixart image it looks a bit more uh over drawing I love the the the the face of the wolf emerging from the darkness of the hood uh both uh models created the concept of the Matrix with The Shining drops and the colors are amazing I think that uh me Journey did a bit of a better job here uh especially in the details of the fur but nevertheless the the image is amazing and beautiful the next image is an illustration of a panoramic view of a mountain Summit with the two-point perspective inward background 80s retro theme blacksmithing depths map sportlight UHD M Journey created a beautiful image and the the resolution of the image the size of the image the dimensions as you can see our portrait uh uh so so the panoramic concept is a bit uh neglected here but it looks beautiful drawing and a beautiful style and this is the image that pixart generated I really like the details here all the lines of the drawing are crisp and you can clearly see the details even on the far away mountains look at the clouds and on the sun I really like uh this variation and I think that pixart here is the winner by far okay so the next image is a porcelain portrait rered using kintsugi kintsugi is a Japanese art where you break something and then you glue it together with a a sort of a glue mixed with a silver or golden powder so this is me journey and without a doubt it generated a beautiful image look at the cracks and and the details of the face it does look a bit like a mannequin more than a humanoid face or more than a human face but without a doubt it's beautiful with all the golden powder smeared on the face and this is the image that was generated by pixart again the details are beautiful it's a bit more uh realistic in the sense of the face but it looks amazing look at the eyes with the eyelashes and look at the golden powder spread all over it looks beautiful I can go on with this forever but let's do one more image now the prom was Picasso painting of extr extraterrestrial alien beings Picasso art classic Picasso this is what mid Journey generated it looks a bit 3D to be a Picasso but nevertheless the image is beautiful the aliens are cute and without a doubt it it passes the The Prompt and this is what was generated with pixart uh I think it it is much more a Picasso style image and it clearly un nailed it when it comes to the prompt we can see extr extraterrestrial alien beings and I think it it's simply beautiful I think pixart is the winner by far in this case but once again you'll be the judge and thank you much everyone for staying this far I hope you enjoy the video hopefully I'll see you in the next video subscribe like and all that see you later bye-bye

Channel: The AI Art

Views: 14,999

Keywords:

Id: KcxssceXNXg

Length: 20min 57sec (1257 seconds)

Published: Tue Dec 12 2023