Probably the Best Model of 2023 So Far.

I have found my new favorite model and it's probably not what you think it is just look at these results here now it's no secret that one of my favorite models for a very long time have been the Juggernaut variants but this isn't it and it has been trained further than Juggernaut and has more input images in my opinion it's even better you know I always strive to get the best realistic images as possible I mean a lot of the models can do pretty colorful vibrant art especially if you prompt for that I think the hardest part is realism and we're getting one step closer oh and I know a bunch of jokes about umbrellas but uh they usually go over people's heads AI these look very nice and I'm also looking at especially here these two are the most close up and if you look at the skin I would never guessed that this was an AI generated image now this is think diffusion XL and it was uploaded today but uh I have had access to this for quite some time so I've been testing it thoroughly now they did send it to me a couple of weeks back so I've been playing with it and full disclosure they have been paying me and sponsoring my videos but this video in particular is not sponsored I actually think that this model is a very good model so there's some comparisons it says the training and images used are over 10,000 they are all hand captioned each image has been tagged by hand to um what actually helps us with with the prompting so if the tagging is done correctly the model will train on those keywords and once we prompt for that it's going to remember the training data and so forth and so forth so the human tagging kind of takes away possible errors that the computer tagging has the potential to do so over 10,000 training images this average model here or others 1,000 to 2,000 and training steps what 1.8 million versus uh roughly 250,000 for for the average model and here are some comparisons easy prompting trained for all art styles and realism 4K data set which is kind of cool most models are trained by especially sdxl 1024 x 1024 this has also a 4K data set now it says here average model does not have some of these things some models actually do so it's not that TD XL only has this but uh like they say it's the average model here that doesn't train on uncensored or not say for workor images human tag training data we talked about that and then the rest is you know every model has it a 1024 data set down no refiner needed well average model everyone except like the base sdxl which requires a refiner but we're going to look at what it actually can do so I've loaded up ruin Focus now which uh I've been using a lot especially to get out just you know simple goodlook images for uh more advanced features I'm probably going to venture into automatic 1111 but I love Focus for its Simplicity let's put in woman closeup portrait in cyberpunk scene raining Neon Lights Let's do let's do sunglasses sunglasses at night like all the cool people do well I didn't put in night actually but we have it just had a picture in my mind of of it being night with with the neon lights and everything so we'll see if it's going to be night or daytime and I'm thinking here that I'm not going to cherry-pick the images for you I'm just going to generate and um you'll see what we get I think that that's going to be a fair way to show show the model I will however speed up the the renders for or your convenience really see here we have some cool looking examples I think this turned out pretty well this one's kind of cool too maybe a little much too pink in the glasses this one wasn't great so well two is turn out well out of those I'm really happy with this one actually even the glasses are kind of symmetric and everything let's try let's change here alien Warrior close-up portraits in sci-fi scene beautiful exotic alien world landscape Battle Scars face painting I have no idea what this will get us I have done a lot of like alien creatures stuff like that I most mostly do humans let's see uh see what happens we're at least looking to get some kind of humanoid here not a lot of face paintings in the first one I am using a cinematic style here so if you prompt for cinematic you are going to get like a more desaturated look which is more prevalent in film uh you might get some film grain prompt for that as well so I know that we can look at some comparisons later but but I I noted when I did some comparisons and when when I prompted one of my usual Styles is Cinematic and try to get some film is at first glance it might look like the images are desaturated in comparisons it actually looks more realistic which is something that I like it looks uh looks a little bit more color graded which would be uh well similar to to high production film so it didn't understand the scars of the face paintings we might want to wake that up a little bit but I I think in general the the character here looks very nice this could be something out of a out of a movie maybe we can try and move that battlecars face paintings put that in in the front let's weight that up a little bit let's also put in colorful vibrant alien world landscape now if you are using automatic 1111 and want similar Styles I have a lot of these Styles implemented also in automatic 1111 the link for those styles are going to be in the description below when you have a cinematic style other their styles that are are fantastic to use and they are all free just check out the link below I think these look cool we didn't get the colors or or the paint paintings that we were prompting for I think the style here is kind of overriding that so I'm going to just drop that out and I'm going to generate again and see if we can um and yeah you can see straight away once we drop that cinematic style that we are getting some more color in these images so so bear in mind that that um instead of trying to waight up and down too much you might just want to you know drop some words in your prompt try to make them them easier or shorter and I've tried some very short prompts and uh it works actually quite well and and as you can see in this one now this is all the only prompt we have we have bcars face paintings alien Warrior Etc and the images are still looking very nice and as you can see here on on on models we are using think diffusion only there is no refiner and there is no Laura attached to this so this is straight out of the box and uh straight away I feel this is much better than the Cinematic aliens we had previously because we were actually prompting for the color and kind of the face painting here wow this is kind of nice even the eyes turned out well and that's something I actually read about so we can try that let's try woman closeup portrait and let's just do blue eyes because I read if you actually prompt for specific eyes instead of getting the the usual sdxl eyes it's going to use the train data set for eyes and and you will get better eyes instead of the usual like marble like eyes that kind of look messed up so let's see if it can uh deliver on on that here I think the first set of eyes here look pretty good we're going to look at the pupils a little soon I am noticing that um well for blue eyes in particular is very is very clear blue so you could try like light blue aquamarine or some sort of other type of blue if you don't want them to be so prevalent in your image and if you are getting images that look very similar to each other you can look at your clip skip value and uh play a little with that so I think um these look very nice and I'm also looking at especially here these two are the most close up and if you look at the skin I would never guessed that this was an AI generated image you can even see here some of the hair is coming from around the mouth here looks looks very nice looks very realistic oh I'm happy with this this looks nice so let's try something else let's try fantasy warrior in epic battle let's add in flowing magic light maybe blue purple teal red I'm going to drop in a digital art style selection here we are going to generate four new images and again I'm not sure picking any images here now the the first alien ones well they were okay but we should have dropped the Cinematic Vibe earlier but let's see what we get here from our flowing magic light of the fantasy warrior in epic battle so the first one here we got some kind of a guy in armor holding a was not a spear I was going to say like sort of a scthe but then it kind of stopped here the second one has a a sword and a dagger but I love the the lights that we are getting here so I wanted this contrast from the blue purple tealish to to the red and we are getting that so far in in two of the images at least the third one seems to be getting that as well and we might be getting a sort of a blue magic spear here so that's kind of cool so now we have a more painterly Vibe this is the this isn't the realism that we had before this face maybe needs a little bit of in painting I mean it's okay from far the hands we missed a little bit but I think the rest of the image the the vibe here is fantastic this looks pretty cool as well even has some depth here with the background now what I would do if I wanted to um improve on something like this is I would take it um send it into automatic 1111 and I would in paint details to to add some detail to the character the face maybe some of the armor that would really make make the image pop a little bit as the cool kids say let's take one of the other prompts here let's take this one here let's just copy that we're going to remove everything going to slap that in and the negative so now we're creating a cinematic film still portrait of a Viking Warrior man with face paintings and blood actually this has let's stop this this has detailed eyes let's go for um to Green Eyes this time oh it's not stopping I'm might need to restart this okay so we have restarted and I am generating again this Viking Warrior with green eyes now this time we have uh not selected a style here we just copy pasted from civi Civ civi what do you say let me know in the comments below I'm truly curious everyone pronounces it differently as you can notice now we are getting like we talked about a more desaturated look because we have a cinematic films still here and even cinemascope so will truly affect the end result of your image but that's something I prefer because you can easily raise the saturation in um my preferences is Photoshop but every if everything is is too much it's hard to you know get it back it's similar to like filming and in log stuff like that so here we have some cool looking images of the Viking Warrior with green eyes yeah I think these are fairly okay really nice actually I like this intense stare that's going on let's try to remove the Cinematic film still kind of just do HDR vibrant collar high contrast closeup portraits let's remove cinemascope and Moody let's see what else is here highly detailed High budget let's remove the Hollywood film there we go and we're doing the same thing with um a little more pulp to the image I hope so let's look at these we are getting some more color I think in general I I might have been taking it a little too far I would probably have liked to seen these with you know less less saturation I think they're a little too much now I'm going to keep using this because you know I like it very lot it's going to be a good base model for me let me know what you think or what your preference is I think my preference previously has been Juggernaut I like that a lot I've also been using realistic stock photos um especially just for basic images so those have been my my my goto models if you try this out let me know what you think if you uh have another model that you think is better that I haven't tried let me know I'll I'll check it out that I did note while um I did some tests of my own but I think this was kind of good kind of showcase here because we have this image here on this is from the think diffusions own page and they have a comparison between some of the models which uh has Juggernaut realistic stock photo here there's also dream shaper and the ex cell base and you can clearly see the think diffusion is less desaturated less glossy than the other ones and here in the text it says uh TDX L provides a very realistic experience without any without an overly saturated plastic feel which is prevalent in for example the stxl base model the colors are inherently more muted for realism but can still be prompted to provide high contrast vibrant colored images while using prompt word cinematic you will truly get a more cinematic vibe to your images and an Exquisite World of Color graded film so that's what I've been trying and playing with and I I think if you prefer that cinematic more realistic style you will enjoy think diffusion as always have a good one see you

Channel: Sebastian Kamph

Views: 43,682

Keywords:

Id: 213L73ojsdk

Length: 14min 16sec (856 seconds)

Published: Mon Oct 23 2023