Hello there engineers and welcome
to this video for Prompt Geek. In this video, I'm going to be
trying something a little different. I'm bringing to you a series now of
regular videos where I'm going to be reviewing the latest models on CivitAI. For those searching for the secret
to realistic AI images in stable diffusion, many have actually held
off on the move from SD 1.5 to SDXL. But could this model be strong enough
for even me to finally make the switch? The model we are looking
at is RealViz XL V2. This came out just six days ago, and
my goodness, the images look fantastic. One of the biggest problems with
Stable Diffusion XL so far has been how plasticky people tend to look. But in this particular model,
they do not look plasticky at all. In fact, we have got some fantastic
grain in the images, some wrinkles and imperfections in the skin. The quality of the images coming out of
this model are really astounding, and of course, with it being SDXL, we are looking
at much higher resolutions as well. So in this video, I'm going to show
you where to download it from... The link in the description down below,
where to put it, and then, of course, the important bit, the settings you're
going to need in order to be able to make this model put out the best images ever. And, of course, some prompts
for you to try out too. So, let's get started. So the first thing to do is, of
course, head over to civitai. com, and I've put the link in
the description down below. It is a hefty file, and that is
because it is an SDXL checkpoint, and they tend to be a little bit larger. Now, this particular checkpoint is a
mixture of several different checkpoints. You can see here that it's a merge of
DreamShaper, Juggernaught SDVN6, and then a couple of other ones there as well. To install it, we just come
along here to Download, now, if you're using Focus, like
I will be, you'll go to Focus. You'll go into the root folder,
down to Models and Checkpoints, and you will save it here. If you're not using Focus and
you are still using Automatic 11. 11 for your SDXL generations you will
actually come through to your Stable Diffusion folder, you will go to Models, and then you will go to the Stable
Diffusion folder to save it here. Either way, you're going to save it. It's going to take a few minutes
because it is a massive file. You can see I've already
downloaded this already, You'll notice though that this
one does have a VAE baked into it. So you'll hit save, let it download,
then you want to start up Focus or Stable Diffusion Automatic 1111,
depending on which one you're using. Now the focus is open, what
we want to do is we want to... Make sure that we are going to use the
right settings for this particular model. A lot of these models do have
best practices, so let's just quickly check what they are. First of all, we do want to use the
correct negative prompts in here. So we'll copy that negative prompt,
come over to Focus, click Advanced, then we're going to paste that here
in the Negative Prompt section. You'll see that the recommended parameters
are 15 to 30 steps using DPM SDE Karras in Focus, with Advanced selected, we
can do that by going to the Advanced tab, going down to Developer Debug
Mode, we can change our sampler method. By clicking on this and choosing DPM
SDE and scheduler will be Karras. We'll want to go to the model tab. We wanna change the base model
here to real vis XL V 20. We don't need the vase. We can turn this refiner off. This LORA is automatically switched on. We can leave it on there. It's an offset, so it will mean
that your images should have a little bit more contrast. And for this particular case, we'll
leave all the other Loras empty as well. In terms of style, we can
leave it on Focus V2 and on Slightly Cinematic if we prefer. And I think it would be better to have
it in more of a portrait style, so I'm going to just change the aspect
ratio to something like 896x1152. So with all the settings ready to go,
we just need to get our prompt in here. So here's my prompt. High Fashion Photography Photo
of a Young Woman with Detailed Skin with Platinum Shignon Hair. Shot on Canon EOS 5D and let's generate
that image and see what result we get./ Now, out of the box, these images do
look nice, however, they do lack a little bit of that kind of realistic edge. You can certainly see the model is
creating some texture on the face, but this does look a bit too polished. Now, part of that could
be that I've actually... said high fashion, and a lot of
fashion photos are very airbrushed. However, there is also the point here
that the actual sampling sharpness could be a little too low on here. So we can come into the advanced tab
and we can actually increase that. Let's put that up to 10. In fact, actually let's take off
the default style as well and see how that affects the image. We'll stick with the same
prompt, but let's run that again. And you can see the result
this time around is a lot more actually realistic looking. It still looks like a
high fashion photograph. It's got that kind of polished look, but
it certainly feels a lot more grounded and real than the last generation. and you can see that here in this image. You can see there you've got the
actual skin textures and The individual strands of hair visible. So let's try that now
with a different prompt. This one's a little bit more involved. It's a large format photo of a woman
with a gaunt face, with black short twin hair, wearing a top, headshot,
edge lighting, high angle, Tri X 400 film in the style of Nan Goldin. And now if you're wondering how I do come
up with these prompts and how I'm getting images that look realistic the answer to
that is available to all of you for free if you use the link in the description
down below and go and download my free E book called creating photorealistic
images with AI using stable diffusion. It has 282 five star ratings
and over 10, 000 downloads. It's 182 pages Packed full of information
about how you can prompt and get amazing results from stable diffusion that
look incredibly realistic All of that for the total price of zero dollars
However, of course if you do want to support the channel and want to support
me Then please do go ahead and donate a couple of dollars towards my coffee
fund, Which is always welcome, because I need that coffee to stay awake. I am working on a new
book for you as we speak. Go ahead, download it for
free, and then get this model downloaded and tested out in SDXL. So, going back to that, let's
generate this next image. The result for this prompt is
even more impressive and you can just see here the level of detail
and realism in those images. This one of course is the closest one
to Tri X 400 because it is a black and white film type and it just looks great. You've got so much detail in the
skin and in the grain of the image. If we have a look at these close
up, maybe a little too much detail. It might be worth even just scaling
back the sharpness on here a tad bit. But as you can see, we
have got a great result. Let's run one more through here. We're going to be going
with this prompt here. Modernist photo of a young woman with
a cute face, with brown Victory Rolls hairstyle, wearing a tailored jacket. Close up on face, edge
lighting, Dutch angle. We've put weight to that. And shot on Aiton LTR, black and
white, in the style of Lee Friedlander. The result, we've got our woman there,
with her victory roll hairstyle, you can see the Dutch angle playing
out in the horizon behind her, it is of course black and white as well
and if we zoom in on this image, you can see that it has a huge amount of detail in
the skin, in the textures of the clothing. We can, of course, then turn this
into an input image to upscale, and we can just drag that down below,
upscale it by two times, and then we will be able to get a much more high
res version of this image as well. After a few minutes, you then have a much
higher resolution version of your image. I just opened this in a new tab. We've gone from 8 9 6 by 1 1 5 2 to
a whopping 1,792 pixels by 2,305. And if we open it up and have a
look, you can see that we have a much higher resolution image. We can really zoom in on this,
and it is very high quality. With huge amounts of detail in this image. I'm going to just do as one last
image, cross process photo of a young woman wearing makeup with
black sleek ponytail hair, blue eyes, headshot, cannon EOS five D. Let's give her actually
something to wear as well. Wearing a sequin jumpsuit. And we should order this correctly,
so let's take the blue eyes because they're more important than the clothes. And we're going to put those here. Okay, and now we're ready to run this
prompt and let's see what we get. And the last one there followed
the prompt really nicely. I would probably say that we can
straight off tell that this eye is off. So, if you wanted to fix that,
you'd go to input image, you'd go to inpaint, you'd drag it in here. You would then paint over the eye, change
your prompt to just a eye with blue pupil, and you'd render to fix that. We're not gonna do that on this
particular occasion 'cause I am just doing this for an example. But as you can see, you can get some
fantastic results with this model. Like I said, using decent prompt structure
like the one I promote in my free book, creating photorealistic images with stable
diffusion, you too can create amazing lifelike images in stable diffusion XL. Now if you liked this type of video,
the review of the different models available on CivitAI, do let me
know by hitting that like button and subscribing to the channel, and I'll
make sure to bring out regular reviews. There's so many of them, and they
take up so much space on your hard drive, you can't waste all your time
downloading them and testing them all out. Leave that for me, and I will make
sure I let you know what the best ones are out there every single week. So, as I said, make sure you
like and subscribe, and I'll see you next time, Prompt Engineers.
