Hey everyone, welcome to Architecture for Thought Today we're going to take a quick
look at some interior and exterior views using Adobe's new Firefly text to
image generator, so let's jump right in When we go to the web page this is what it looks
like, and at the bottom of the screen is where you enter your prompt, so I have one ready and loaded,
I'll just paste in and while it's generating I'll read a little bit of it off it's,
"A highly detailed prospective rendering of a modern minimalist simple kitchen
with natural oak cabinets" etc. etc. I will post all of these prompts in the
video description, but right off the bat you'll notice the images come in very clean
looking very similar to stable diffusion There are some artifacts
that are not exactly precise, but for the most part they're pretty good images So I'll start just by changing this to a 16x9,
and I did another video where I described how all of these tools on the right hand side here
work, so I'm not going to go too deep into that, but you have the option to change it to appear
more like a photo, more like artwork, etc. etc. I'm just going to switch this out to a "Photo"
for this series and then down below change some of the color settings to be a little more vibrant,
the lighting to be a little more golden, and the composition to be a wide angle shot and I'm going
to rerun this one more time and then we'll get into how to use these images as reference images
to continue developing further in this program So it's a pretty interesting process here
that adobe is is giving you the ability to do I like this image here so I'm going
to click on these three dots here and it gives you the option with the pull
down menu to "Use as a reference image" and as soon as you click on that it's
going to regenerate these images again using that image as a reference image, and you can
see it dropped it down below here just off to the side of the prompt, and it gives you this nice
little slider bar where you can make the image, or the new generations of images, more like
the reference image or more like your prompt So again once it comes out using the reference
image it's not going to be too different, but you can test the image generations
by playing with the little slider bar to create new images or stay more
closely related to the reference image So you can see here just an example of that, so
we're going to do something interesting here, we're going to... I actually liked the
reference image, so I'm going to pull it back to where we were a little bit, and
I've got some other prompts that I'm going to grab that are completely different, and
we're going to see how it treats basically new generations with a new
prompt, and the reference image So I'm posting this in here and I'll
read a little bit of it off it's "A hyper realistic cinematic photo of a
futuristic flowing minimal interior living space terracotta clay arches tall
intricate ornamental ceilings" etc. etc. So you can see right off the bat the composition
of the image stays relatively the same meaning the perspective view, you're even getting
a lot of the geometry that's the same in terms of the cabinet layout, and the beams
running across the ceiling with the windows, so now let's pull that slider bar even more
towards the prompt just to see basically what this would be producing if it were the prompt only,
and as it's doing that I do want to jump into, basically to explain what I think is going on
in the background here it looks like it's using what's called controlnet, which takes the either
canny edges of the image or uses a depth map to hold the geometry of the original image, and so in
this example here I'm using stable diffusion with canny edges through its new controlnet kind of
input feature, and the original images off to the left here and then I typed over it an "all black
elegant interior", it uses the candy edges or a depth map image to hold the geometry and then it
basically reproduces the image with the new prompt So that's what it looks like is going on here, and then it's even giving you basically a
slider bar to hold closer to the reference image through those canny edges or depth
maps or go closer to the prompt so... These are pretty interesting here I think
let's do one more, I'm kind of just tweaking it a little bit more towards the reference
image, and then we're going to take it a next step further and bring it back more towards a
realistic image, so yeah these are excellent, I like this option here so I'm gonna
use this as a new reference image, it's going to create some new ones,
but I'm going to go grab another prompt And we'll paste this in after these generate Very nice, but let's go ahead and do this
next one, which is a "boutique retail shop interior minimal trendy cool all white fluffy
couch and seating furry white rug" etc. etc. So we're pretty far over towards the right hand
side of the reference image bar, but you'll see if I go back towards the middle it's going
to hold some of these arch geometry columns, and even in the background you might see some
of the columns starting to appear back there, so these are turning out pretty,
pretty great, pretty interesting I might even want to go a touch more towards the reference image just to give it that real,
kind of surreal type of look and environment So pretty neat, so you can jump in here once
you are given access and play around with these, I'll download one of these images and pull
it up onto the screen just so you can see the fidelity and the detail in the image So I kind of like this one down here, let's
just download this it pops up with a little prompt telling you that it's going
to stamp it with some credentials, so basically it lets everyone know that
you created this image in their program They're trying to be really good about declaring
you know what content is being made using artificial intelligence, and I think
that's a great thing the transparency, so you can see the detail in the image
from the prompt that we created and holding a lot of the geometry from some of
those other images that we started from so pretty neat and then I had another example
basically I wanted to show you of what that controlnet uses in the background
the depth map or the canny edges That's basically what was used for a
previous video I did on how to make some of these images move or rotate, it uses that
depth map image to basically give you some foreground elements background elements,
but again hold the geometry in the image so those are our interior views next we're
going to jump to some exterior views, and I have a prompt that I'm going
to load in first we're going to clear all of our stuff out here And while it's generating you'll
see basically what it would have generated without any prompt so very
different from reference images so you can see the effect that it had so
pasting in our exterior and this is a "flowing curved small villa sculptural large
glass windows glossy white stucco and matte black leather" and these are some fun images that
we got I did a whole series on these basically I'll just run through 20 - 30 of these images
where I cut the slider bar right in the middle and basically just did a bunch of generations
over and over to produce a series like this where you can see the foreground always
has the swimming pool there's always the rolling green hills in the background, but
the geometry is slightly changing every time Which is fun and great to produce you know 20,
30, 100 options, but again you need to know what you're looking for to actually use these tools a
little more specifically so this is a great image down here we'll use this as a reference image and
it's going to generate some new images based on that so as you are developing it I've noticed
that if you really tried to take out all of the prompt as you're prompting over this it's going
to not do a great job of reproducing the image So you need to make small tweaks on your
prompt if you want to actually control the geometry that is coming out of the generator so
here I'm pulling the slider bar back a little bit to give it a little more freedom, both of
these images are really cool, kind of like, well this one is actually pretty great, let's use
this as a reference image, and then let's change some of these words, or let's take a bunch of this
out just to show you what what happens basically So we're getting some pretty similar results but
if I just take a bunch of that out and we change maybe "flowing" to... "orthogonal" ...spell that right? And generate it with the prompt slider bar pretty
close to the prompt we can move it a little more again to see what it starts to create, but
it might you might start getting some more rigid geometry, but you can see immediately it's
starting to produce a different type of of image And as I continue to pull the slider
bar more towards the prompt you're going to get less and less of that reference image So that's basically what I
wanted to show you guys today Hopefully it's a useful tool, and I'm really
curious to see what you guys are starting to generate on either programs like this or other
workflows, so feel free to leave me some comments, I really enjoy going back and forward to
push the use of artificial intelligence in architecture and I think it's going
to be sort of a group effort to elevate these technologies to where they'll actually
be, become a useful tool in our profession So thanks for tuning in, and
I'll catch you next time, bye!
