Hey everyone, Welcome to Architecture for Thought Today we're going to take
a quick look at image to 3D Yes it exists It's amazing how quick these
AI tools are developing, and I do believe they have a place in the
architectural discipline especially in early concept phase designs where we may
be able to 3D print these digital meshes But also potentially for final production
full-scale architectural 3D prints We're going to focus on a program called
ZoeDepth on the Hugging Face website Then jump into some research called
Symmetry Driven 3D Reconstruction And lastly we'll take a sneak peek at a one-stop
shop AI driven 3D software called 3D Guru Let's jump right in So per usual I will leave a link in the video
description to all these websites, but we will start with the ZoeDepth on the Hugging Face,
and when you come here you will see three tabs "Depth Prediction" "Image to 3D" and "360 Panorama" We're going to focus on these last two tabs today,
and I've got three examples that I want to show We're gonna do three different scales,
an architectural sized scale building, an architectural detail, and then the 360 panorama So, let's start with the building, which this image was generated in
Midjourney, so an AI generated image You'll click submit after you upload it, and
I have the "Keep Occlusion Edges" unselected And that is the edges that basically
connect the very distinct foreground from background will not include geometry,
so you can see in this model off to the right hand side here that there's no geometry
connecting the background to the actual model This is looking pretty good in
this little image viewer here, but let's download the model and take a
look at what it looks like in a 3D program So this is looking pretty great We've got a lot of depth on
this geometry and you can see the mesh is fairly accurate it to the image I can definitely imagine some programs
in the future, AI driven, giving us the capability to basically rotate these images
based on the depth maps and in these meshes, and then further being able to prompt, you know
from this perspective to continue basically the geometry or the style of architecture over
on the back side here to complete the model You can imagine even something like DALL-E's
in-painting being able to erase and kind of regenerate parts of the model, but I'm
going to rotate the sun here so you can see you know how much shadowing and
depth there actually is on this geometry And uh yeah, this is this is pretty impressive
already to even have a model to this level of geometry, and again this is a digital mesh
that we can 3D print if it were closed on the back side so even if you brought this into
blender or some other kind of sculpting program, and you did include the occlusion edges which
would connect the foreground to the background you may be able to pinch part of it off,
and actually 3D print this as a relief model So very interesting, I'd be interested
to see how you guys are planning to use this tool and some of your own research,
but let's jump back into our next example So next we are going to load in a architectural
detail, and this is for a 3D printed chair It's a node or a connector, and I have the
actual model of this so I was interested to see how accurate this program would
generate it, and it's looking pretty good Let's download this and take a
look at the model in our 3D program So this is looking pretty great, I'm going to
rotate the model here so you can see the geometry, and it did a pretty good job of accurately
portraying the geometry from the image I'm familiar with what this node actually
looks like, and it's got the curvature, it even looks like it's got some of the smooth
versus some of the hard edges, pretty accurate, so again even with a model like this you
know being able to use a program like this in conjunction with the ability to upload
another image, or being able to text prompt to generate more image for content aware 3D
geometry production, you can imagine being able to complete this model and then exporting
it as an STL and being able to 3D print it as a prototype or as a final production model for
actual use in product design or architecture So this is looking really great Let's jump into our next example Okay, so last but not least we're gonna
click on the 360 Panorama to 3D tab and it gives you a little disclaimer
there basically saying that uh it's not trained to produce images or 3D
depth maps based on spherical data, but we're going to take a look at what
we can do with the data that it produces So for this one I'm going to
actually keep the occlusion edges on, and I'll show you why in a second It'll produce basically what looks like this
little bubble here, and it's a little strange it kind of flips it on its side so once you
export it you may have to rotate it flat, and I'll show you that in a second, but before we jump
into that model I want to show you the actual 3D virtual reality image, which is
right here, we'll do full screen And you can see you know these
are very high quality images, but the point I wanted to make is
that it's only from one perspective, and we're basically aligned with this edge of
this wall here, and the wine cooler, and on center with basically the range and the island and
as I rotate around the fireplace on the other side So let's jump into the 3D model and we'll
take a look at how we can actually rotate or move ourselves through the space
to gain some additional perspectives All right so I've loaded this into my
3D Builder and I've rotated it flat, and you can see again it's just
basically a bubble from the exterior It's wrapped in the panoramic image, but let's
jump into this bubble and see what we can see So now I'm on the inside, and when I
was describing before basically that the panoramic image that we saw in our company's
website we were aligned with basically this wall and the wine cooler, and when we rotate on
center with this range, this basically because it's a 3D model gives us the ability to
actually shift out of center with that, and because we kept the occlusion edges
on, surfaces that are distinctly broken for instance such as the island, it
actually keeps that geometry as a surface cutting all the way back to
the next depth plane that it has so This is really amazing, I mean it's
producing basically the 3D model from that panoramic image and
even over here if you can see the edge of the island you know you're
seeing it move in front and behind Basically you know the spaces around it So this is really amazing, I mean it gives
you a little more freedom in the space basically to explore panoramic images from
a panoramic image that you already have made And again you can imagine how you can start
to link some of these images or be able to come in here and text prompt over parts of
this image, or you know connect some of these bubbles together to get actual 3D models
with interpolated geometry between them So this is an amazing tool, another
kind of great advancement in the AI image to 3D models that we've got going on here, very curious to see what you guys are able to
come up with using some of these technologies Alright, next I wanted to take a quick
look at this research called "Symmetry Driven 3D Reconstruction from Concept Sketches" And I'll leave a link in the description
to this PDF if you want to look through it, but about halfway down it gives some
examples about how they're using the symmetry based on these concept
drawings to produce 3D models, and they actually released a video with some
examples of it, and I'll play a clip from here And it shows the sketch in the top left
corner, and then the reproduction of that sketch in 3D based upon an axis basically,
and there's a couple variations of it, but we can definitely see how a
program like this mixed with the depth map and the digital meshes that we're
just looking at in some of the other programs, can be combined to give you the capability to
actually edit the 3D model with sketch lines So this is another research program called
"Sketch2Mesh" and it's reconstructing and editing 3D shapes from sketches, but then it looks
like actually gives you the ability to redraw, for instance it's showing a line being
drawn on the top here and then adapting that 3D generated model to fit that new line,
and so they give a couple examples of that here I wanted to show this because I think it's pretty
interesting to imagine basically that you will be able to create these 3D models, rotate them, and
draw and adjust the model based on your scribbles, or sketches, or text prompts to complete
the models or just refine them based on how you want them, and then you basically
have a 3D print ready model to work from So last but not least as I mentioned
at the beginning of this video I had a conversation with the developer of
this program, which is called 3D Guru A very interesting program and
there's a video of it on their website here that basically shows you
can text prompt the size of a room It makes a room for you in 3D You can text prompt furniture pieces and it
will create those furniture pieces in the space You can text prompt images or textures to apply
to the wall, ceiling, floor, etc. and use those as inspiration or reference images for then
text prompting an image in the same program So this is all-in-one, basically a one-stop
shop, where you can do all of these things and he showed me a tutorial, they gave me access
to it, I'm not going to do that in this video, we'll do another video where I kind
of go through that in more depth But it's really interesting to see that
we're starting to get 3D programs now, not just these Hugging Face
websites or anything like that Where we have these capabilities, AI capabilities, in the 3D programs to give us the ability to
render, produce content, models, place the models, you know generate room sizes, different
types of rooms, just based on text prompts So I wanted to give you guys a
glimpse at this because I've seen a couple other programs as well
starting to do similar things One was called Spline, and I'm signing
up for all these waitlists basically to try and get in so I can show you guys what
these things look like in real time and... I do realize these technologies
are in their very early stages, and what I'm presenting here is very experimental If you're still in school, or just out of school, and just learning the traditional workflows
of architecture and the architectural process, I still encourage you to do that because it's
important to understand the difference in the way things work now, and will continue to
work for probably the next five to ten years, or even beyond that, before we start to transition
to using these tools primarily in the office The more I work with these tools the more I'm
starting to see programs like Revit start to feel like AutoCAD because they are so progressive
and when combined with the ability to control parameters, and edit in 3D based on text prompts,
and just quick generative designs and variations, you can see how quickly we'll be able to jump to a point where architecture does become
more "Raster" based meaning image based rather than these precise vector line drawings
that we've been doing for a very long time That is representation, and this is presentation
of the actual content that will be immediately 3D printed or digitally fabricated, so again I
understand these are very experimental workflows, and I want to encourage you to still pursue
the traditional workflows so that you can understand if you are attempting to implement any
of these technologies what the differences are, and why you would do that in one
scenario versus another scenario So that's going to be it for now, I did want to
mention I had the opportunity to be interviewed by Dimitar Pouchnikov who has a YouTube
channel called the UH Studio Design Academy He does amazing Blender tutorials, and has
a great outlook on architecture in general, so I'll leave a link to his page in
my description, you should go over there and follow him so when the podcast
does come out you'll be able to watch it We had a great conversation, basically
about what I was just mentioning, the importance of knowing the traditional
workflows and still 3D modeling digital geometry, and where the transition might start to occur
with the implementation of AI in architecture So thanks again for coming by,
and we'll see you next time