Greetings, everyone. In this video, I am going to share with you
a one-click downloader to download very best Stable Diffusion 1.5 based custom models into
your computer or into RunPod or into any cloud. I will show both RunPod and Windows how you
can download all of these amazing very best Stable Diffusion 1.5 based models. Moreover, I have generated their X/Y/Z checkpoint
comparison with 9 images per prompt. You see a photo of a man. This image is over 1 gigapixel. I am going to make a detailed analysis of
each model for 768 to 768 pixels resolution. I am also going to make a detailed comparison
of the image of a man prompt and I will explain the difference between the image of a man
and photo of a man prompt for 768 to 768 resolution. I am also going to compare 1024 to 1024 pixel
resolution and show you which one of the models are able to generate accurate images, good
anatomy having images for 1024 to 1024 pixel resolution. I am also sharing all of these X/Y/Z plot
comparisons. You see a 1.1 gigabyte file, 1.1 gigabyte
file, 1.2 gigabyte file, 1.5 gigabyte file, 2 gigabyte file. You can see their grid resolutions 1.2 gigapixels,
1.5 gigapixels, 2.1 gigapixels. During the video, my comments will give you
a lot of tips and you will see which models are best for stylization or realism or cartoon
or 3D. This video will be amazing. The reason why I have generated such file
and I am making this video is that I will pick the very best realism model and I will
do empirical experimentation on that model to find out the very best DreamBooth training
hyperparameters with Kohya and you will be able to use those parameters in OneTrainer
or in other trainers as well, which will help you to train your Stable Diffusion 1.5 based
models with amazing quality and accuracy and generalization. I have done the same thing for SDXL base version. It is working amazing. I have done over 100 trainings for SDXL base
version. Hopefully, I will also make the same for SD
1.5 based version. So this tutorial is recorded for that purpose
and this tutorial will show you the very best models for every category for Stable Diffusion
1.5 based models. So before start comparing the X/Y/Z plot comparisons
of all of the 161 models, I will show you how you can one-click download all of them
into your Automatic1111 Web UI folder on Windows and also on RunPod and use them if you wish. The link to this Patreon post will be in the
description of the video. You don't have to be my Patreon supporter
to follow this tutorial and see the which models are performing best because I will
show them on the screen for you. Still, I have also uploaded the files, original
files into this post, so you can also download them onto your computer and compare them later
if you wish. To download all of the models with one click,
just download this .bat file and .py file. For Windows, these two are just sufficient. Go to downloads, edit the .py file and in
here, there is download models. You will see at the very bottom, you see download
folder here. Just change this folder wherever you want
to download. Let's say I want to download inside my G drive
inside temp7 folder wherever you want. So just replace this folder path like this
and save it and then go back to where you have downloaded files. Double-click download top 160 SD models .bat
file. Run anyway, it will run this .py file and
it will ask you how many simultaneous downloads you want. I suggest you use one simultaneous download
because if you use more threads to download, you will likely to encounter errors and you
will be needed to rerun the script. Let's say three, for example, it will start
three simultaneous downloads. So what is good about this script is that
let's say there were some errors during the download. After the script run has been completed, rerun
the script and it will only download the not accurately downloaded files. So it has a continue feature. You see sometimes you may get errors while
downloading. The script will handle everything automatically
for you. Just run the script again and it will download
all of the files fully accurately. We can also see that the download has been
started inside our given folder. You see Realistic Vision version 1.1, Realistic
Vision version 6, Majimix. Let me also show you all of the models that
have been downloaded previously. All of the models are here. Let me show like extra-large icon. Let me show like as a list. Here you see all of the models and you may
be wondering how many gigabytes they are. When I see the properties, you see 411 gigabytes
is the total size of the models. So if you don't want to download all of the
models, just comment the models like this that you don't want to download. And the script will not download the models
that you don't want them to get downloaded. So how can you download them on RunPod? First, let's log in to our RunPod account. So from this link, you can go to their sign-up
or login. So this is my account. Let's go to pods. Let's click new pod. The interface slightly changed, you see. So in here, first select your template. You see choose template. I will use Stable Diffusion template. Wow, where is the search? They changed the search. So I will search with the browser. Okay, RunPod Stable Diffusion Web UI template,
this one. Then you can use either secure or community
cloud. I will use community cloud extreme speed. And let's go with RTX 3090 deploy customized
deployments set the disk volume persistent disk volume according to how many models you
want to do or not. Let's set overrides continue deploy. Let's go to my pods. Let's wait until it is loaded. I have amazing RunPod tutorials on our channel. Just click this search icon. You see the search icon. And just type Runpod. Or whatever the search term you are looking
for, you will get all the videos that I have for RunPod with this link. So if you don't know how to use RunPod, you
can watch these amazing tutorials. Okay, let's connect to the Jupyter lab from
here, upload the .py file, nothing else, which is here, change the path if you need. Since we have changed the default path, I
will re-set it, which is at the very top here. So the accurate path will be inside models,
Stable Diffusion folder, let's copy the path, paste it like this. Okay workspace like this. Yes, it is accurate. Then just open a new terminal. You see currently I am not in the workspace. So let's move into the workspace like this
and just type Python and download top 160 SD models. Let's download with five threads and it will
start downloading everything into the accurate Automatic1111 web UI folder. Again, if you get errors, run the script one
more time until you get all the models fully downloaded. That's it. So it is downloading with 100 megabytes per
second speed. I have spent quite a bit of time to prepare
this downloader over one day. Now we can begin testing the results and finding
out which model is best. I will delete this pod. Until you delete your pod with terminate,
it will remain as it is. So I will just terminate. Okay, so all the images are here. How can you download them? Right-click, click Save link as and select
the download folder. Let's download into downloads. You can download every one of them. And you see the resolutions are here. So if your computer is not strong enough,
you can download the half-resolution of the grids or the 25% resolution of the grids. But since my computer is strong, I will download
the full-resolution files. So let's just download them. The file size and resolutions are all written
here. This is the way you download. Okay, I have all the files downloaded here,
obviously. So how am I going to open them? I will use the Windows default photo viewer. So let's begin with a photo of a man. I did test three prompts for 768 to 768: photo
of a man, image of a man, and picture of a man because photo of a man will add realism
to the generated images. Therefore, I tested image of a man and picture
of a man too. So you can see which models are more realistic
and which models are performing for more like anime or cartoon or 3D. Okay, let's begin with a photo of a man. Currently, my aim is finding the very best
model for realism. Just wait until the image fully loads because
the images are huge. Therefore, it will take time to load. You see still not fully loaded. Okay, now fully loaded. Let's go to the very beginning. This is 768 to 768. This resolution is currently almost as a standard
for SD 1.5 based models. But I also have tested 768 to 1024 and 1024
to 1024. So we will be able to see best performing
models at high resolution. Okay, the names of the models are written
at the top so you can see the first model. This is the second model not very good. It generates a lot of black and white here
as well. Absolute reality. All right. Abyss orange model, aing diffusion model like
this you see Airtist MAL Realistic version 2.5 model. In some of the models, you will see almost
the same man because the majority of the models are just merges, not new trainings. Therefore, they will print the same man. Okay, this is Airtist MAL model like this. This is AmIReal model. You see there are some major problems with
this model. This is Amour Bold model. This is a pretty good model. You can see the quality here. It's a pretty decent model. And this is just photo of man, nothing else
in the prompt. No negatives, nothing else, CFG 7. And I have used DPM++ 2M SDE Karras sampler. This is the best sampler that I find and use
it for the steps. Okay, this is Analog Madness model. You see pretty decent quality. But about diversity, are we getting diverse
faces? No, the faces are very similar, but pretty
good quality. And this is AniMerge model as you are seeing
right now. Here's the model. It is not very realistic. This is AniMics model. You see it's a pretty decent quality, but
at distance shots, of course, the faces are not that great. This is AniToon model. Oh, the face is really good at a distant shot
too, as you're seeing right now. So this is AnReal version 3 model. Again, I see all blue eyes. These models are overfit. This is an AnRealSpiceMix as you're seeing
right now. This is AnyLoRA model. Not that great quality. This is Any LoRA Anime Mix model. Not that great quality either, but it has
an anime look. That's for sure. This is Art Universe version 8 model. This is how it is. This is Arthemy Comics version 6 model. Okay, this has a problematic output. Not that a great model. This is Arthemy Real version 1. Again, I almost see the same face as the other
models. This is AWPortrait model. Not that a great quality. Probably Asian model. This is Beauty By Stable Yogi version 3 model. Not that a great quality. This is Beauty Fool Ultra Real version 1 model. Okay. Not that a great quality either. This is BeenYou model. It looks like Asian anime mixing model. Not that great quality. Beyond Reality version 2 model. Not that great quality, but the faces are
pretty different and that's a good thing. If you are getting pretty different faces,
that is a good thing. That means that the model is not very over-trained. So this model, Beyond Reality version 2, has
some decent training. That's for sure. This model is one of the good ones. This is Blazing Drive model. This is the style that it has. This is a blue pencil. I don't know why I am getting like this output. This is Cheese Daddy model. Not a good one. One of the old models and you see it is bad
outputting. This is Chik Mix model. Almost same man. Not a good model. Colorful version 8 model. Not the quality is not very good. This is Consistent Factor Euclid model and
I see the almost same face again. This is Corbe Realism model. I don't know this model outputs a very skinny
man like this and the almost same man every time you see. So pretty overfit model. This is Cornucopia version 2.2. This model has a very very good clothing and
environment quality. It is like a cinematic shots. The faces are very similar but this is also
another looking good model. The quality of the environment and the pictures
and the clothing are really good. This is Counterfeit version 3. An old model. Not that good. This is Cyber Realistic 2.5D. Not that good. This Cyber Realistic Classic model. Yeah, almost those repeating same man images. Cyber Realistic version 4. Again the quality is like this. Not that very realistic. Cyber Realistic version 4.1. Almost same, you see there aren't almost any
difference between version 4 and version 4.1. All right, let's continue. Daring Hydra model. The images are like this. This has a somewhat chubby and that is good
because it's a diversity in the output and therefore this is also a decent model. It prints not exactly the same face or body. Okay, this is Dark Sushi 2.5 model. This is Dark Sun version 7B model. As you are seeing right now, this is Dark
Sushi Mix version 2.25. By the way, if you are also a model developer,
you can message me and hopefully I will add them to my downloader. My downloader downloads will be counted on
your model downloads. Don't worry about that. And this is Deliberate version 3. An old model but I still included it. This is the output. This is Deliberate version 5. You see not that great quality if you ask
me. This is Disney Pixar Cartoon Type A model
and this is the output. It gives not a great quality obviously. This is Divine Anime Mix Version 2 model. Not that great. Divine A Legend Mix. By the way, I noticed that usually the new
anime models have higher base resolution. This is Dream Shaper Version 8. The realism is not good, but it has a stylized
output and it is decent. This is Dreamy Cybergasm model. You see, I see the same face as in the others,
so another merged model I can say. This is DucHaiten Retro. This looks like a little bit like Asian model,
and also I noticed a noise here, so this is probably one of the older models that also
got affected from some training bugs. Epic PhotoGasm Last Unicorn model. Okay, this model is one of the really good
models, you will see that the images are really high quality, realistic, not exactly the same
face, some difference, they have some different persons and looks, so this is one of the really
good models, Last Unicorn. There is also a Epic PhotoGasm Z Universal
model, let's compare both of them like this, pretty similar models, not exactly the same,
so you can use the Epic PhotoGasm latest model, it is a pretty decent high quality model,
the resolution of this model is also good I will show you. This is Epic Realism Natural Release Candidate
1 model, the face is almost the same person, but the realism is really good, you see the
skin tones, the skin textures, the details, this is another really good model. The only problem is that it prints almost
the same face, so it is pretty overfit, but it is a really realistic model. The details and the realism is really, really
high. This is Ether Real Mix model, you see, like
this as you are seeing right now. This is Experience Version 10 model, it has
a certain style, not that realistic. This is FaeTastic Version 2 model, this is
an old model probably not a good quality. This is Fantastic Mix K1 model, this also
has the noise that older models have. This is Fenn Foto FF2 model, not a good one,
you see. This is Flat 2D Animerge, this model is really
good if you want to turn your images into 2D anime type. It is pretty strong overfit model for 2D,
but it works good, and the quality is also decent, you see. Okay, this is Ghost Mix model, not that good
one. This is Gui Realism Photo Version 1, I almost
see the same face as others, probably another mixed model. This is Guo Feng 3 Version 3 model, this looks
like an Asian model but not that great quality. This is Hello Art Oil model, you see, it has
an oil style output, it's decent. This is Hello Kid 2.5D Version 1 model, the
quality is somewhat decent. This is Henmix Real Version 4 model, this
model also doesn't look that good. This is Himawari Mix model, you see, this
is its style. This is Hyper Realism Version 3 model, okay,
it has different faces in all images, that is really, really good, but the quality is
not that, I don't know, quality is also looking decent. So this is one another realism model, it looks
pretty decent and pretty accurate too, it also looks like very, very diverse output,
you see, every face is a different person and that is a very big plus, Hyper Realism
Version 3. Okay, again another model, you see, this model
when we compare these two models, the right model Hyper Natural is extremely overfit,
the face is very similar, but when you look at Hyper Realism Version 3, the faces are
completely different at each image, therefore we can say that Hyper Realism Version 3 is
much more generalized model than the Hyper Natural Version 1. Okay, this is another model, ICBINP SECO,
you see, almost the same face in every image, it is a decent quality but not much generalization. This is Incredible World Version 4, this looks
like a somewhat not realistic output. Yeah. Jernau Mix Chimera, you see, almost the same
face in every image or like the other models. Juggernaut Aftermath, okay, Juggernaut is
very popular but I don't find it as realistic as the other models that we have seen, it's
pretty decent quality but you see the face is very similar, it is not like a different
person in each image. This is Juggernaut Final model, again decent
quality but almost the same person, same pose. Juggernaut Reborn, you see these models are
really overfit compared to the other model that we have seen. This is JXKRE model, not a very good one. This is Kawaii Realistic European model, okay,
this model also has noise. This is Korean Style 2.5D model, not a good
quality. This is Kotos Mix Version 1 model, not a good
quality. This is Lahmix Mysterious Version 2, it's
a decent quality model, you see, but not that extremely good, but decent. This is LEOSAM Film Grain model, and this
is the output it gives. Yeah, I don't think so that's very good. Okay, this is Limitless Vision Version 3 model,
you see the outputs like this, not that a great model. This is Lofi Version 4, not that a great model. This is Lyriel Version 1 model, it has a certain
style, and not that a great model. This is Mag Mix Version 9, by the way, when
I say not that a great model in terms of realism, I am saying that, so these models can be good
for other tasks, other prompts, but I am just comparing for the most realistic model right
now for studio shot-like quality that we can do DreamBooth training on. Okay, this is another model, this is a decent
model for like an Asian model, you can see Mag Mix Version 9. This is Majimix Alpha Version 2 model, the
quality is decent, only problem is I see that this face is very similar in each image, and
there's also some anatomical problem, like on this image, yeah, anatomical problem. This is Majimix Fantasy model, you see this
is the output. This is Majimix Lux model, these are the output. This is Majimix Realistic model, not that
realistic, that's for sure. This is Manmaru Mix Version 3 model, you see
this is a really, really good, really, really high-quality cartoon model, I can say that,
Manmaru Mix Version 3. This is Meina Mix Version 1, this is also
pretty decent model for its style. This is Meina Pastel Version 6 model, let's
compare both of them, there are some just like minor differences, I don't know. This is Meina Unreal, this is also almost
the same, you see. This is Milky Chicken model, yeah, this is
the quality, not that great one actually, decent but not, not sure. Yeah, this is Mistoon Anime model, this is
a really good anime model, you see, this is almost like from animes, pretty decent quality. This is Never Ending Dream model, not that
a good one. Next Photo Version 3, okay, this looks like
a realistic model, but the quality is not there, and I see some noisy images as well,
you see, there are some really big noise. It has different body shapes and somewhat
different faces, but it also has a noise. Okay, this is Non Plus Ultra Version 2 model,
okay, this also has some noise. This is Noo Sphere Version 4, it has a certain
style and really good quality, but the style is like this, if you like this style, that's
a good one. This is Noo Sphere Version 2, almost same,
you see when I zoom in like this. Okay, this is Objective Reality model, it
is like this. This is Only Realistic Version 3 model, it
is like this, not very realistic. This is Opiate Version 3 model, the images
are like this, it's a decent style model, but not that very much realistic. This is Orange Chill Mix model, this is Paragon
Version 1 model, you are seeing right now, almost same face. This is Pastel Mix, if you want this style,
this is the model. This is Photographer Alpha, let's look at
the quality, the quality is decent, but almost same pose and same person. This is Photon Version 1 model, and this also
looks like pretty much previous models, this is the quality it has. This is Photo Somnia Omega model, the output
is like this, you see, almost same face, I don't know how many times we have seen like
this man. This is PicX Real model, and it also looks
like another merged model, this is its Version 1. Okay, this is Porous Rox Version 1, it has
a really good style, if you are into this style, it's a good one. It looks like a little bit over-trained, for
example, almost same man. This is PVC Style model, and the output of
this model is like this, you see, if you are looking for this kind of output, pretty good
one. This is QGO Prompting Real Version 1, not
that different from other models. This is Raemu Mix Version 8, and this is its
style. This is Raemu Mix Version 4, you see, this
is Real Dream Version 12, looking like other models, not much different. Real Cartoon 3D, not like a real cartoon,
so this is its output. Real Cartoon Anime, okay, this looks more
like an anime. This is Real Cartoon Pixar, okay, this is
a really good one. So, if you are looking for a Pixar model,
Real Cartoon Pixar is really good. You see, the quality is decent, it looks like
a Pixar like 3D rendered, this is a good model for this task, and remember, this is just
photo of a man, nothing else in the prompt. This is Real Cartoon Realistic model, okay,
this is somewhat okay, not very realistic but decent quality. Yeah, okay, this is Real Dos Mix, doesn't
look like a good model. This is Realisian Version 5 model, not much
like an Asian, but yeah, it's up to you. This is Realistic Version 5.1, okay, there
is some overfitting, some bad quality. Realistic Vision is known to be best for realism,
and it's really good, but I think that the previous model were better than this one. Okay, this is Realistic Digital Genius model,
like this. This is Realistic Fantasy Version 2, like
this. This is Realistic Version 6, okay, Realistic
Version 6 and Realistic Version 5.1 has some major difference, you see, Realistic Vision
Version 6 has a huge amount of new training, that's obvious. Let's look at the Realistic Version 6, now
this is good, this is really good, the only problem is the face is almost the same, not
much diversity, but the quality is really decent. Still, it's a choice, but this is looking
pretty good, but I think I will prefer the model that outputs different person in each
image because that model is the most generalized model. This is Realistic Asians Version 7 model,
it is decent, but not that great. This is Real Life Version 4, this looks like
another merged model. This is Rev Animated Version 1.2.2, this model
is not continuing anymore, but is a decent model for that style. This is RPG Version 5, this looks like biased
to older persons, but it's decent, you see, these are the images that it outputs. This is Run Diffusion FX model, so if you
want to train an older person, probably RPG Version 5 will be a good choice, I think. So this is Run Diffusion FX Photorealistic,
let's see the quality, I see some noise, yes, this model also has some noise, so I wouldn't
prefer it. This is Sals Realism, this is an extremely
overfit model for like this fitness guys, so this is not a generalized model. Okay, this is Samaritan 3D Cartoon model,
this is a really good model for this style, I like it, a little bit looks like over trained,
but if you are looking for this style, it is pretty decent. Okay, this is SDVN3 Real Art model, and you
see, this is its output. This is SDXL Unstable Diffusers Version 11
model, this is its name, but this is SD 1.5 based model, and this is the output it has. This is Semi-Realistic Version 54, let's see
the quality, not much like different, you see, this is not a good one. This is Serenity Version 2, it's also looking
like the other models, a merged model probably. This is Soft Realistic Version 155, it also
looks like another merged model. Okay, this is Stable Diffusion Version 1.5,
which is the base model of all models that you are seeing here, therefore it is a very
bad quality, this is the official model. This is Stoked Reality Version 2.2 model,
you see the output is like this, not that great. This is Summershard Version 1, and it is pretty
decent for this quality, this is a stylized model. This is SXZ Luma 0.9X model, you see, this
is its output, not that great. This is T3 2.5d Version 1 model, and not a
great model. This is 3T version 10 model. Not a great one either. This probably an Asian model. This is the Truality Engine Pro model, the
quality is like this, it is also looking like another merged model. This is TMND Mix SP Rainbow model, this is
a decent model for this style, if you are looking for that. This is TMND Mix Version 7 model, it is like
this. Okay, this is Tooneame Version 1 model, and
it is pretty decent quality too. So this is another like toon models. This is ToonYou model, this has a certain
style, and it's a pretty decent good quality model, as you are seeing right now. This is Umbra Mix model, and this looks like
another merged model, almost the same face. This is Unyielding Grimoire model, you see
it has a certain style, but it looks like one of the models that we have seen. This is 1.5 pruned CKPT, this is also the
base model. Okay, this is Vendo Photorealistic Version
2 model, and it doesn't look like that realistic. This is Virtuos Mix Version 1 model, it has
a certain style like this. This is Wein Comic Version 1 model, and it
is very decent for this style, if you are looking for such style. This is Wildcard X Version 2 model, looks
like another merged model. This is Wyvern Mix Version 9 model, this looks
like a differently trained model, not like a very merged model. If you are going to need people like here,
I think this is a decent model for such style. Okay, this is XX Mix 9 Realistic model, and
the output is like this, not very good. This is YACAM SR16111 model, this looks like
another merged model. This is Yayoi Mix Version 2 model, not that
a great model either. This is Yes Mix model, it is not looking good. This is Yazora Version 1 model, since it's
a decent model, I think it is for this style. And this is AniVerse Version 1 model, and
it's also a pretty decent model, and the author contacted me and told me that he is also going
to release another one soon, and he said that it will be even better. Only this image is problematic, but the other
ones are really decent quality. So I am not going to also show all of the
picture and image difference, but if you want to see which models are more stylized, then
you should try picture and image to see because the prompt makes a difference. When you use photo word, it will make the
images more realistic. So if you are going to compare stylized models,
you will get much better results when you type image of a man instead of photo of a
man. So this is the difference, I put them into
the Patreon post, you can download them and check them out. You see, when I type image, the stylized models
become much, much better because the photo prompt will make them more like a realistic
output, but when you make them image, you will see the stylized models will get much
better outputs. So this is very crucial when you want to test
stylized models versus realistic models. For realistic models, you should test photo
word; for stylized models, you should test image or picture word. You see, this is another image, this is another
one, Raemu Mix. So the stylized models will shine and look
much better when you do image or picture prompt. This is a really good tip for you, this is
what I have discovered. You see, when I type image, it is much better
in this Samaritan model, so the image of a man will perform much better for stylized
3D anime models. That is the difference, this is ToonYou, you
see the faces are different. So this is a really, really important thing
when you consider. Okay, as a last step, I am going to compare
1024 to 1024, this is double the size of the base resolution of SD 1.5 models. However, if a model is already capable of
higher resolution, that means that model has more potential than the other models, but
I will make my choice for SD 1.5 based on 768 to 768 when I am searching for the best
hyperparameters for DreamBooth training. However, if a model has a base higher resolution,
then it is better for us. So what happens when you generate images in
a higher resolution? You will see repeating subject in the images,
like here I typed photo of a man, but I got two men here or two men here, so they are
repeating, or you will see the anatomy is broken like this, you see this is incorrect
anatomy, incorrect anatomy, almost all images are incorrect anatomy. So the majority of the models cannot handle,
as you are seeing right now, 1024 pixel images. I noticed that the anime models are much better
at high resolution, such as this one, but the realistic models are usually generating
very bad anatomy at 1024 to 1024, you see, the anatomy is not that great, or here, the
anime models are performing really good, but they still also getting some errors. Okay, you see bad anatomy, and let's check
it out, another model, it also doesn't look very accurate. For example, this face, anime models performing
good. So in here, which model shines for the very
best resolution? Let's check it out quickly and I will tell
you. For example, this is also very bad one, and
this is also bad one, the merge models will all produce bad quality. This is also another bad one. Okay, this is also another. Okay, yeah, this, some images are decent,
but some are bad ones. This also become very bad, you see, two persons,
very bad anatomy, a lot of major errors. This is also like that, major errors. This is also very bad looking. And let's see if there is any realistic model
that can output at 1024 to 1024. You see, major errors. This will happen when you go with a very high
resolution. This was a decent model, but at 1024, it is
also not very good. The anime models are significantly performing
better. This is also very bad. You see, this is also bad. Very bad. Okay, this is also bad anatomy. You see, this was a bad anatomy. There are some multiple subjects, bad anatomy. Okay, this performed really, really bad. Good. This is Dark Sun model. This is another model. Uh, it's looking decent. This is also erroneous, you see. Deliberate. Deliberate, let's see if there is a very good
performing. This is also bad. Duc Haiten Retro. Okay, this model significantly performs good
for this resolution. Duc Haiten, but the images are not like that
we are looking for, and it also has some, you see, this noise. Okay, this is Epic PhotoGasm, which is one
of the good models, and it is still performing decent. For example, the anatomy here is bad, but
I think in this image, this one is looking nice, and this one also looking nice. So in some images, the anatomy is broken,
but in some images, it is still good. So this has a really high resolution having
model. Okay, this is Epic Realism Natural model,
and the anatomy is not that great. Okay, there are also some multiple subjects
in this model. This model, let's see. Experience Version 10, it also decently handled. And this is FaeTastic model, this is Fantastic. By the way, I am moving very fast, so just
pause the video while watching and check out the parts that you want because I have to
be fast otherwise this will take hours, there are just too many models. Okay, this is Gui Realism Photo Version 1
model, this Guofeng model, and it came like this at this resolution. This is Hello Art Oil, wow, this model performed
really good at this resolution, interesting. This is Hello Kid 25D, and it really handled
the resolution very well. So this is also another high-resolution model. This has multiple subjects. This is Himawari model. Okay, this model, Hyperrealism Version 3,
performed excellent at this quality. Every image is almost a different person,
you see, it's printed out this resolution very, very well, no major anatomy. This model is epic, I can say that. Hyperrealism Version 3 is really, really realistic,
as you are seeing right now, I like it, it handled this resolution perfect. Okay, you see, this is also another merged
model, and these merged models are epically failing, as you are seeing right now. Okay, this model also performed very well. So if you are looking for this kind of style,
this model is really good. Incredible World Version 4, it really handled
well this resolution as well. This is also another fail model, another fail
model, another fail model, another fail model. This resolution is really high, you see, these
models are all failing. Okay, this is another decent model. Okay, this is another decent... Oh, anatomy is not that great. This is Korean style, not much Korean left
at this resolution. Kotos Mix, okay, like this, some major errors. Okay, Lahmix Mysterious, this model is really
also performed well for this resolution. So this is a really, really good model again. Okay, this is another model, Asian models
are not that good actually, none of them is performing really good. This is another model, I see some anatomical
problems. Lofi, you see, anatomical problems, a lot,
anatomical problems. Mag Mix Version 9, let's see, yes, this model
produces noise at this resolution, a lot of noise. Majimix Alpha, okay, yeah, anatomical problems. Majimix Fantasy, anatomical problems. Majimix Lux, okay, this looks like performed
the best. Majimix Realistic Version 7, anatomical problems. Manmaru Mix, Manmaru Mix also looking decent,
I see some anatomical problems but decent. Meina Mix, like this, Meina Mix also handled
this resolution excellent. So it's a good one. And this is Meina Pastel, okay, this model,
Meina Unreal, looking also decent. Okay , this is also not a good one. Milky Chicken, Mistune Anime performed really
good at this resolution. Never Ending Dream, like this. Next Photo, not that great. Non-Plus Ultra, you see the same merged model. Okay, Noo Sphere also handled this resolution
excellent. Yeah, really good one. Noo Sphere Version 4, Noo Sphere Real Version
2, this has a certain style, you see, I don't like it very much. Objective Reality, not that good. Only Realistic Version, not that good. Opiate Version 3, not that good. Orange Chill Mix, not that good. Paragon Version 1, decent. This is a decent model. I see some duplication here. This happens when you go very high resolution,
but no anatomy problems. So, this is another good model. Paragon Version 1 for high resolution. Pastel Mix, like this. Photographer Alpha, decently. Not that very decent actually. Photon Version 1, like this. Photon Version 1 is also, let's see, yeah,
somewhat okay. Another, Photosomnia, you see, this is bad. Not that great. Pic X Real, I see anatomy bad. This is anatomy bad. These are all merged models. Okay, this is also anatomical bad at this,
Porous Rox. This is a high resolution though. This is not the native resolution. This is double size of native resolution. This is PVC Style. This is QGO Prompting. This is Raemu Mix. Okay, this is Raesan Mix. Not that great. This is Real Dream. This is Real Cartoon 3D. Real Cartoon 3D also
performed really good. This is Real Cartoon Anime. Okay, this is Real Cartoon Pixar, and it really
handled good. You see, almost this is its native resolution. This is amazing. Real Cartoon Pixar Version 8, this is the
best Pixar model probably. Real Cartoon Realistic. Real Dos Mix. Realisian Version 5, not that great. This is also performing bad at this resolution. Realistic Vision. Realistic Genius, I see some anatomical problems. Realistic Fantasy Version 2, anatomy not that
great, but it performed decent. Realistic Version 6, let's see. Yes, the author really increased the resolution
of the model with Realistic Vision Version 6. You see Realistic Vision Version 5 fails epically
for anatomy, but Realistic Vision Version 6 handles anatomy, but the problem is almost
same person in every image. You see, it is overfit. Realistic Asians, yeah, here like this, not
that great. This is another merged model, it fails. Rev Animated, wow, RevAnimated really performed
well for this resolution. This model was epic, but it is not continuing. Okay, this is all double subject, so this
resolution was too big for this model. Run Diffusion, not that great. I see some major noise in every image, you
see. Noise, noise. I don't know why they have it. This model also all duplicated. Sals Realism, Samaritan also couldn't handle
this resolution. Anatomy fully broken. SDXL based version, you see all different
images like these ones, but since this is a very basic prompt, it is handling like this. Refiner couldn't handle this resolution, interestingly. This is SDVN3 Real Art, like this. So many major errors. This is SDXL Unstable Diffusers, it also handled
pretty decent this resolution. Okay, this is Semi-Realistic Version 54, decently
handled, I can say that. This is Serenity Version 2, using anatomy
broken. This is Soft Realistic Version 5, anatomy
broken. This is SD1.5 model, it is all now noise at
this level. Stoked Reality, anatomy broken. Summer Shard, uh, it is like this. SXZ Luma, wow, this model also decently handled
this resolution. Okay, T3 model, it is like this, not a great
model, if you ask me. T3 V10, okay, anatomy broken. The Truality Engine, TMND models like this,
it is really handling this resolution too. Tooneame, this is also performed really good
at this resolution, it is good too. ToonYou also handled this resolution really,
really good. These are really good models. Umbra Mix, it also looks like handled, but
not that very accurate anatomy, you see. Unyielding Grimoire also looks like handled
this resolution pretty decent. This is base model, Video Photorealistic,
no, the anatomy is broken. Virtuos Mix, not that great model. Wein Comic Version 1, wow, this model also
handled this resolution really, really good, you see. Wildcard Version 2, anatomy broken. Wyvern Mix, anatomy broken. XX Mix 9, it also looks like handled this
resolution, but not a great for either style or for realism. Okay, anatomy broken YACAM. Yayoi Mix, anatomy broken. Yes Mix, this is its style, decent. Yazora, this is its style, and Yazora really
performed well for this resolution. And AniVerse Version 1 also handled this resolution
almost perfect, really, really good. This model also has a really good potential. I will hopefully also update this file and
add my findings as a text, so you will also see them. I hope you have enjoyed, hopefully see you
in another amazing model. And as a next step, what I'm going to do is,
I will pick the very best model here and do empirical experimentation to find the best
hyperparameters for training SD 1.5 based models. Hopefully see you in another amazing tutorial
and thank you so much for supporting me.
