There's been an explosion of AI-created images in the past couple of months on social media. Using midjourney, Dall-E and other AI blackboxes, designers and artists alike have generated a flood of images. The creative process goes something like this:
1) Craft a text prompt to communicate a visual idea to the blackbox
2) Adjust various parameters to toggle the generated image
3) Post-process image
As the author points out, these black boxes do not create drawings or details. They build images out of an ocean of prior references - a supercharged but cruder version of collaging and hacking, that rough initial design process that turns nebulous ideas into concrete concepts. To the extent that designers also survey references and precedents to find manifestations of abstract ideas - and starchitects delegate watercolor dreams to junior employees to convert to concept drawings - these black boxes complement or supplement these human processes.
I'm personally a bit disturbed by how easily some designers have ceded that rough process to the blackboxes, at least for social media projects i.e. 2D images. Would students simply type prompts into these blackboxes in lieu of collaging and charette-ing? What of creative ownership when the designer knows nothing about what goes on in the blackbox?
I have been playing with these for a bit now and the ratio of good to trash is low. There's a ton of garbage for every randomly interesting image. You can see some of my dall-e mini (craiyon) here: https://archinect.com/forum/th...
Craiyon is the worst of them, in my opinion. The last one is disco diffusion, which is the most interesting, but also the hardest to get anything useful from.
To answer your specific question: I see a use for these same as any other inspirational image, if you take what you get and use it literally, then there's low creativity and thought in your process and, while it may LOOK cool, it's not going to be very satisfying.
I have a few acquaintances who are taking over their Instagram’s generating this sort of imagery. Often using their own work as a steppingstone, and generating new images that they are claiming semi-authorship of. The images generated are often very reminiscent of the things I haven’t liked coming out of the field for some time. Ephemeral, chaotic, functionless, but still often beautiful.
At the same time, given that I work on projects being built in the real world, I don’t actually care all that much. If this becomes more ingrained in the process for designing real buildings, great. I’d be all for this technology being leveraged to generate design iteration, or even renderings of actual designs (neither of which appear to be in it's capabilities yet). The latter being something I don’t like doing the leg work for anyway, and that I think is often wasting the abilities of the people doing it.
I am curious how this might impact idea’s competitions. 10-15 year ago, If someone developed something like this and didn’t share it with the public, they could probably have smoked a bunch of competitions and made bank. Curious what it does now.
I was involved in some research in grad school with respect to working with AI and algorithms to produce design iteration and be a helpful tool for architects during the design process. My personal views are that the image generation side of this is a neat party trick, but probably has little practical application unless it's simply to get people to break outside of the box in terms of how they might think of image as architecture, or reimagine these images without preconceived notions or baggage. Even then, still more of a thought provoking exercise rather than something that could be directly useful as part of a finished design deliverable.
More interesting to me are the types of applications that take input of constraints and generate design iterations based on that and evaluate them against a fitness function of project criteria. I think the promise there is more efficient use of the architect's time to evaluate those iterations and adapt them to less easily defined, or even subjective, constraints or criteria (e.g. aesthetics). I see some tools come across my timeline from time to time being developed as plugins to Revit/Rhino/etc. that are getting pretty interesting.
I haven't seen a big push to involve these tools too much in design, but it is happening to some extent. It's not as flashy, and it probably doesn't have as broad appeal as showcased on someone's instagram feed for example, but I see them as likely part of the profession moving forward much more so than the AI-generated images as renderings I've been seeing more and more of lately in those feeds.
One of the early design studios I took in college was doing something similar (mashing up other context/images) in a more analog and (hopefully thoughtfully) considered way to generate inspirational images that then served as a jumping off point for our studio work that semester. These tools would have been an interesting way to do essentially the same ... but probably would have driven that particular professor crazy with the implications. Some of the other studios were doing similar but with collaging found objects and trash in physical models, collaging images in photoshop and other digital tools, etc. Perhaps I'm not opening my mind enough, but I sort of see these image generating bots as something similar. Something useful in a second-year-studio type of way to help drive conceptual design, but not really that applicable in actual practice.
OK, so AI can now do what any first year architecture student with Sketchup or Rhino can do...yawn. When it can do a constructible set of CDs, let me know.
the construction set is probably the easiest to automate. Though I would not worry overly much about it happening so soon.
The creative stuff at the beginning can certainly be inspired by the AI stuff I am seeing around the social media feed lately. So many beautiful images and a lot of it simply inspiring. So far I have not been able to get anything like as nice as some of my colleagues have, but it does feel like a useful tool to get outside of my/your/our head and see things afresh. If this is the beginning then I cant wait for the next iterations of this software. Its brilliant.
Some of the drawings Matias (del Campo) is doing are quite beautiful. He warns people that there is a lot more work involved. A lot more than just writing prompts. Probably you've seen some of them Will. They have been circulating on his FaceBook page.
Gorgeous yet unsettling. It begs the question - could any human artist have created these images without the blackbox's assistance? Could anyone have imagined these images at all? The common cultural ocean (Made by generations of collective human creativity) from which they are drawn is available to humans too but there's something beyond success via brute force in these creations.
I'm reminded of the Roald Dahl short story, The Great Automatic Grammatizor, where an engineer realizes that all human stories follow rules that are largely mathematical. He invents a machine that is able to out-write human authors. But the machine doesn't eliminate these writers - rather, the engineer corners the market by persuading out-matched but famous authors to put their names to the superior literature his machine generates by the dozens everyday ...
Aug 16, 22 5:20 pm ·
·
SneakyPete
That reminds me a bit of The Nine Billion Names of God.
Aug 16, 22 5:25 pm ·
·
atelier nobody
It is pretty. Not sure real world economics would allow it to ever actually be built, but I'd definitely play a computer game with that as a setting.
Like anything this popular, it will hit the mini-malls like the post modern and deconstructionist architecture did. That is my prediction with the exception of a few image makers for the hi-end markets.
Midjourney has already been overrun with images of celebrity faces popped onto busty cosplay fantasies. For every interesting prompt there's 3 prompts that will be used carnally.
It's like a lot of things, perhaps an AI may produce an image, or some prose, or a building. A poor representation of those tied to an artificial form of architecture whom are most at risk.
The biggest problem with this culturally and historically is that all the images are based on western anthologies, cultures, philosophies, art, language, technologies, and more. This is a problem of further colonialist development paired with AI. It draws all its sources and resolutions, speculations from already established knowledge of the west, it is not a unifier but unless this area is addressed, you will get results based on a certain richness but not the other(s). Think about, 99% of what we read, learn, and teach. Where are they coming from and what they are based on?
This is an important aspect. Even if the image set is inclusive, though, it will be missing critical understanding that should come with, resulting in yet another set of creations that make images without understanding what those images mean to the people they were taken from.
Yes, and that's the digital erasure of very valuable knowledge from the earth. Capitalism is great at data-based hegemony and mass influence. Maintenance of humanities isn't a big concern.
All very true, but I would also encourage you to look at work being done by folks that are aware of this limitation and trying to leverage the medium to tell different stories in spite of this problem. In no way is it "the solution," but it is a cognizant start.
Artificial intelligence can now make convincing images of buildings. Is that a good thing?
https://www.archpaper.com/2022...
There's been an explosion of AI-created images in the past couple of months on social media. Using midjourney, Dall-E and other AI blackboxes, designers and artists alike have generated a flood of images. The creative process goes something like this:
1) Craft a text prompt to communicate a visual idea to the blackbox
2) Adjust various parameters to toggle the generated image
3) Post-process image
As the author points out, these black boxes do not create drawings or details. They build images out of an ocean of prior references - a supercharged but cruder version of collaging and hacking, that rough initial design process that turns nebulous ideas into concrete concepts. To the extent that designers also survey references and precedents to find manifestations of abstract ideas - and starchitects delegate watercolor dreams to junior employees to convert to concept drawings - these black boxes complement or supplement these human processes.
I'm personally a bit disturbed by how easily some designers have ceded that rough process to the blackboxes, at least for social media projects i.e. 2D images. Would students simply type prompts into these blackboxes in lieu of collaging and charette-ing? What of creative ownership when the designer knows nothing about what goes on in the blackbox?
I have been playing with these for a bit now and the ratio of good to trash is low. There's a ton of garbage for every randomly interesting image. You can see some of my dall-e mini (craiyon) here: https://archinect.com/forum/th...
Craiyon is the worst of them, in my opinion. The last one is disco diffusion, which is the most interesting, but also the hardest to get anything useful from.
To answer your specific question: I see a use for these same as any other inspirational image, if you take what you get and use it literally, then there's low creativity and thought in your process and, while it may LOOK cool, it's not going to be very satisfying.
I’ve been thinking about these.
I have a few acquaintances who are taking over their Instagram’s generating this sort of imagery. Often using their own work as a steppingstone, and generating new images that they are claiming semi-authorship of. The images generated are often very reminiscent of the things I haven’t liked coming out of the field for some time. Ephemeral, chaotic, functionless, but still often beautiful.
At the same time, given that I work on projects being built in the real world, I don’t actually care all that much. If this becomes more ingrained in the process for designing real buildings, great. I’d be all for this technology being leveraged to generate design iteration, or even renderings of actual designs (neither of which appear to be in it's capabilities yet). The latter being something I don’t like doing the leg work for anyway, and that I think is often wasting the abilities of the people doing it.
I am curious how this might impact idea’s competitions. 10-15 year ago, If someone developed something like this and didn’t share it with the public, they could probably have smoked a bunch of competitions and made bank. Curious what it does now.
I was involved in some research in grad school with respect to working with AI and algorithms to produce design iteration and be a helpful tool for architects during the design process. My personal views are that the image generation side of this is a neat party trick, but probably has little practical application unless it's simply to get people to break outside of the box in terms of how they might think of image as architecture, or reimagine these images without preconceived notions or baggage. Even then, still more of a thought provoking exercise rather than something that could be directly useful as part of a finished design deliverable.
More interesting to me are the types of applications that take input of constraints and generate design iterations based on that and evaluate them against a fitness function of project criteria. I think the promise there is more efficient use of the architect's time to evaluate those iterations and adapt them to less easily defined, or even subjective, constraints or criteria (e.g. aesthetics). I see some tools come across my timeline from time to time being developed as plugins to Revit/Rhino/etc. that are getting pretty interesting.
I haven't seen a big push to involve these tools too much in design, but it is happening to some extent. It's not as flashy, and it probably doesn't have as broad appeal as showcased on someone's instagram feed for example, but I see them as likely part of the profession moving forward much more so than the AI-generated images as renderings I've been seeing more and more of lately in those feeds.
One of the early design studios I took in college was doing something similar (mashing up other context/images) in a more analog and (hopefully thoughtfully) considered way to generate inspirational images that then served as a jumping off point for our studio work that semester. These tools would have been an interesting way to do essentially the same ... but probably would have driven that particular professor crazy with the implications. Some of the other studios were doing similar but with collaging found objects and trash in physical models, collaging images in photoshop and other digital tools, etc. Perhaps I'm not opening my mind enough, but I sort of see these image generating bots as something similar. Something useful in a second-year-studio type of way to help drive conceptual design, but not really that applicable in actual practice.
Have nothing to add discussion-wise, but this was an interesting architecture-related video on the topic:
OK, so AI can now do what any first year architecture student with Sketchup or Rhino can do...yawn. When it can do a constructible set of CDs, let me know.
the construction set is probably the easiest to automate. Though I would not worry overly much about it happening so soon.
The creative stuff at the beginning can certainly be inspired by the AI stuff I am seeing around the social media feed lately. So many beautiful images and a lot of it simply inspiring. So far I have not been able to get anything like as nice as some of my colleagues have, but it does feel like a useful tool to get outside of my/your/our head and see things afresh. If this is the beginning then I cant wait for the next iterations of this software. Its brilliant.
Some of the drawings Matias (del Campo) is doing are quite beautiful. He warns people that there is a lot more work involved. A lot more than just writing prompts. Probably you've seen some of them Will. They have been circulating on his FaceBook page.
yes, those ones are what I was thinking of. Also the images by Andrew Kudless (matsysdesign), and a few others, are very striking.
Image generated by Matias del Campo
Victor Horta inspired?
Looks like a 3D printed Gaudi
Gorgeous yet unsettling. It begs the question - could any human artist have created these images without the blackbox's assistance? Could anyone have imagined these images at all? The common cultural ocean (Made by generations of collective human creativity) from which they are drawn is available to humans too but there's something beyond success via brute force in these creations.
I'm reminded of the Roald Dahl short story, The Great Automatic Grammatizor, where an engineer realizes that all human stories follow rules that are largely mathematical. He invents a machine that is able to out-write human authors. But the machine doesn't eliminate these writers - rather, the engineer corners the market by persuading out-matched but famous authors to put their names to the superior literature his machine generates by the dozens everyday ...
That reminds me a bit of The Nine Billion Names of God.
It is pretty. Not sure real world economics would allow it to ever actually be built, but I'd definitely play a computer game with that as a setting.
Like anything this popular, it will hit the mini-malls like the post modern and deconstructionist architecture did. That is my prediction with the exception of a few image makers for the hi-end markets.
Midjourney has already been overrun with images of celebrity faces popped onto busty cosplay fantasies. For every interesting prompt there's 3 prompts that will be used carnally.
The end- of soft serve season- is close…
Bob Ross is not of this earth.
Not news…
It's like a lot of things, perhaps an AI may produce an image, or some prose, or a building. A poor representation of those tied to an artificial form of architecture whom are most at risk.
I just see it as a last grasp of the tech bro's.
The biggest problem with this culturally and historically is that all the images are based on western anthologies, cultures, philosophies, art, language, technologies, and more. This is a problem of further colonialist development paired with AI. It draws all its sources and resolutions, speculations from already established knowledge of the west, it is not a unifier but unless this area is addressed, you will get results based on a certain richness but not the other(s).
Think about, 99% of what we read, learn, and teach. Where are they coming from and what they are based on?
*think about from which 'data' these engines are generating images.
This is an important aspect. Even if the image set is inclusive, though, it will be missing critical understanding that should come with, resulting in yet another set of creations that make images without understanding what those images mean to the people they were taken from.
Yes, and that's the digital erasure of very valuable knowledge from the earth. Capitalism is great at data-based hegemony and mass influence. Maintenance of humanities isn't a big concern.
All very true, but I would also encourage you to look at work being done by folks that are aware of this limitation and trying to leverage the medium to tell different stories in spite of this problem. In no way is it "the solution," but it is a cognizant start.
Can you post some examples? I'm interested.
Check Olalekan Jeyifous‘s post on instagram. Great stuff.
*posts
Block this user
Are you sure you want to block this user and hide all related comments throughout the site?
Archinect
This is your first comment on Archinect. Your comment will be visible once approved.