AI in Arch Viz

[Francisco Penaloza]

Just wanted to see how many people still a live over there and know how it is going with all this AI 'revolution'

I have been having many problems  setting an opinion, it changes all the time, sometimes is good, other is very pessimistic. I have been learning a few tools applying some workflow to my daily routine. but I don't want to say much yet, to leave the conversation open, I would like to know how you guys are doing.

Clients are asking for it?? partners are showing this new shinny thing that can do wonders, or it is just meh...

it would be also interesting to know how long you been doing Arch Viz, kind to balance opinions on experience I guess. I am old times, about 20 years and counting.

Cheers!

 

[Scott Schroeder]

Look into the Gartner Hype-Cycle. We're quickly approaching the peak of inflated expectations and heading into the trough of disillusionment. We too are facing a few leadership people claiming "This is the end of Photoshopping people as we know it!" Everything is going to be super-duper easy. Then they get into Stable Diffusion, spend 6 hours getting prompts right, then another 8 hours correcting the 15 finger hands, and somehow still think the current AI generative fill options are somehow faster than inserting high-quality cut out people. It's architecture, so they are always looking for the next easy button only to reverse course in a few years because they've lost their look and everything they produce looks like everyone else that is constantly seeking the same easy button.

The funny thing is that AI, in the current form, is a lot like Enscape. It has a distinctive look to it, mostly hyper abuse of the orange-teal color scheme. So it's incredibly clear who is using AI and of course, everything looks the same.

Where AI is becoming useful is generating options for exterior views and quickly changing seasons or lighting conditions. I've found it incredibly useful to create mood boards to then go back into the old-timey way of rendering to produce much higher fidelity visuals than AI currently can. This is where AI is going to break a ton of ground but it will not replace traditional visualization anytime soon. The studios that are seeking to use this as a flat out replacement will find themselves far behind the studios that understand that AI is a useful addition to the toolkit, but you are not going to use it as a sole replacement for any one process.

I'm entering my 21st year doing visualization. Like the grizzled old salt smoking a cigar in the back of a dimly lit room, I've seen them come and I've seen them go. AI is going to stick around with us for a while and I am looking forward to integrating it into my workflow, at least until the lawsuits start getting ruled on and completely change the landscape of the current models of AI. If Stable Diffusion were to lose against Getty, that will set AI back years. I don't see it happening, but the laws are a little gray on how they can use the data scrapped from the internet and copyrights.

[James Vella]

Have to agree with Scott on most points. The only things ill add to that are people replacement is actually fantastic. You have to mask the people out or do a separate pass without them so it makes it easier for composite but the basic idea is fill your scene with anima people, swap them out with realistic generated AI people and knock your renders up a few notches, add some rain/splash etc. You can see the comparison before/after here on this thread. It was showcased by a guy called Matt Hallett Visual on facebook.

https://forum.corona-renderer.com/index.php?topic=39728.msg214150#msg214150

 

Also the last few years I have been automating everything I can with python/maxscript and chatgpt is very helpful when understanding difficult concepts. While its far from perfect, actually breaks 99% of maxscript that it writes (especially lately) it can certainly give you insight into where there is a problem in your script or teach you things which may be difficult to understand in the documentation (like having a free tutor on standby 24/7). So my take away from AI at its current point is similar to what Scott said - its a tool that can be useful in the right hands as long as you understand its not always correct (even though it states things as fact), images produced by it can have artifacts or a specific look depending on what its referencing, copyright infringement is certainly a possibility since it doesn't reveal correct references and basically its just a large library of content which its trained on which means anyone who decides to change that content or feed it incorrect data will skew the results one way or another. While it has a bright future (especially some of the tools coming to Photoshop and the likes) at the current point - take it with a grain of salt and be aware of the consequences of where AI can take you or your business. Read the EULA so you don't get yourself in any trouble when using it for commercial work or just be savvy.

 

I've been in the industry 15 years, I embrace changes as any artist that reduce my time in front of the screen. I still think we have a long way to go since large shifts in technology like this usually take about 5 years to mature and another 5 to integrate throughout but within that 10 year span I think it will certainly change things on nearly every creative front for the better. It could happen faster but I wouldn't worry about losing work to AI, probably the opposite - in my view it will just take away the non-creative jobs like when we used to paint out wires or clean up green screen mattes, painfully boring and should be done by the computer. Leave the creative work to the humans.

[Scott Schroeder]

I purchased Matt's tutorial series and while it is indeed a game changer, the AI 15 finger issue persists for me. Over the years we have a fully curated library of cutout people ready to go, so the time it takes for me to fix AI's issues with the human form is still much too long compared to pulling from a great library of ready made cutouts.

However, I don't think the AI human form issue will last for long. I think the next batch of learning models that it pulls from will help solve what hands are supposed to look like. The repose plugin for Stable Diffusion in Photoshop is crazy helpfull though. As you said, put 3D people in your scene and no matter how goofy their pose is you can repose and rework them using AI to get excellent results.

Edited August 3, 2023 by Scott Schroeder
can't spell for shit

[Francisco Penaloza]

Interesting input guys, thanks. I am also a in House Viz person, and I absolute relate to what Scott mentioned about Enscape, actually I did introduce Enscape on our company, we were looking for a solution to replace the craziness that it was with all the different studios using different renders, such, REVIT and V-Ray, Lumion, Twinotion, Cinema 4D. man people doing crazy stuff.  Enscape fit the bill for us. and funny enough some people questioned me saying why are you teaching us a software that will replace you... well I knew that any Architect or designer or their most wisdom... they didn't know how thing work really LOL. after several month lot of people uses Enscape for everything, but they all complain that the image look the same, what button I push to make it look different.

Regarding AI, my company is having a very cautious adoption because, copyright issues, and I do agree with that. Even the phrase of 'using AI to design this project' is, a very conflicting sentence, because.. yea the law are none existent for that yet. and besides, the tech is only doing graphic options, is not following codes or rulles, not even common seance decisions. really is not AI yet.

To me the main conflict I have now is not on the technology it self, but, the use and propaganda that it is around it. You know here in USA the good enough is pretty big. that's how Visualization got cut in half during recession times. it didn't matter those overseas rendering were half or less quality than the ones produced here, but for the price they were good enough. I see the same thinking for lots of people to happens again. I know that the low level project will suffer from that first, but there are people when they see some one else doing it, they ask, can we do that too?  What you guys think.

BTW this tech is moving so fast, that almost at the same time Matt Hallett was posting his workflow, I was doing the same on my own, and it was pretty cool finding other people using AI from a very artistic and technical way, other than just doing pretty anime girls.

Here are tow of my post. links here

I guess I don't know how to share a post from Facebook.... I am old.

 

Placing people on rendering is a real pain, so I am more than happy on using AI to do that, even I been using it to refine Landscaping, using same principal and it work pretty good too.

 

Edited August 3, 2023 by Francisco Penaloza

[James Vella]

On 8/3/2023 at 10:59 PM, Francisco Penaloza said:

Even the phrase of 'using AI to design this project' is, a very conflicting sentence, because.. yea the law are none existent for that yet.

There are strict laws around this already, which is why Adobe only license AI content based on IP they own. Same goes for everyone else, if you question the IP then its not yours and its liable for lawsuit in any commercial work. I mean sure people use Miele appliances in their renders all the time and not get questioned, but its no question that its 'illegal' so they could pull the rug at any time if they choose to. Until then as long as its not defamatory they probably don't care as its free advertisement, even though the artist didn't ask for permission, which again is copyright infringement at the end of the day. 

On 8/3/2023 at 10:59 PM, Francisco Penaloza said:

I know that the low level project will suffer from that first, but there are people when they see some one else doing it, they ask, can we do that too?  What you guys think.

I don't understand your question?

If you are referring to cost cutting, then in my opinion that is usually related to the kind of work or client you have. For example real-estate is dead cheap, previous years we had professional photographers going on-site taking photos, interior designers styling the sets with beautiful furniture and accessories to sell the estate. This still happens in some cases but when it comes to cheap apartments and the like then a smart phone and an appointment is all you need. Thus drives the prices down for the marketing content. "Should we send someone out to take photos? how much? $300", "Should we get someone to render it in 3D? how much? $800, no thanks go take the photos with your smart phone". That's business 101.

When you work in real-estate you work at the bottom level, I've fought hard over $500 per render and now I don't bother, its not worth it. When I work with developers (people who are selling 500x houses, parks, shopping centers etc.) they don't flinch at a bill of $50,000. Do a few aerials, some nice marketing videos/animations, etc. and your studio can pocket $50,000 in a month no problem. Its all about your client base in my opinion, if I understood you correctly.

On 8/3/2023 at 10:59 PM, Francisco Penaloza said:

BTW this tech is moving so fast, that almost at the same time Matt Hallett was posting his workflow, I was doing the same on my own, and it was pretty cool finding other people using AI from a very artistic and technical way

I looked at your AI replacement, very good job!

Yes you are correct, most of us feel the same pains, some do something about it and some don't, its those who don't that get replaced by AI. Those who do something I call 'creative artists', always forward thinking based on a common set of problems or limitations.

[Francisco Penaloza]

On 8/5/2023 at 11:17 AM, James Vella said:

There are strict laws around this already, which is why Adobe only license AI content based on IP they own. Same goes for everyone else, if you question the IP then its not yours and its liable for lawsuit in any commercial work. I mean sure people use Miele appliances in their renders all the time and not get questioned, but its no question that its 'illegal' so they could pull the rug at any time if they choose to. Until then as long as its not defamatory they probably don't care as its free advertisement, even though the artist didn't ask for permission, which again is copyright infringement at the end of the day.

Well yes, my point was Intellectual property. As it is now, there is not law that protect AI generated IP, things gets a little confusing because AI used open data from the internet to 'learn' what it does, and that is what is being questioned now. Not what it being created. From what I read, the USA law won't let you copyright or own anything that you created using AI, you can show case it, you can sell it, but you don't own it. Now if some one else sees that you used part of their work to create something that you are selling, then you are in troubles, and that's why Adobe is re building their datasets with their images( well not their images but the one people give them the right to use them)

So if any architect shows a project and say, I designed this building using AI ( that really until now is not possible because all the tools a very limited) then, they don't own that design. The law only protect anything created by humans. Funny thing is, you can copy right the text that you used to generate the image, but since AI generators are random, then the image is not your IP.

I think when they really nail the way they will monetize AI tool then the law will come up. As for now is the wild west. That's why Stability AI is not charging for their tool, but since Adobe sells software then they are trying to avoid any possible issues. Mid journey is the one that can go down. But since their company has create such Hype about their tools now they can be too big to fall when a 'final law' come up. Which it would be totally none fair.

Talking about ownership, do you know that there is a dataset for stable diffusion that is called MIR. As you may think, yes is a dataset created using renderings from the studio MIR that has a very distinctive style. MIR website is very minimal, so I can't corroborate this information.  But if this wasn't created by MIR studio, can you imagine that in a very close future, anybody could create something that looks like your work, that is beyond traditional inspiration.

 

 

 

[Rodolfo Tanno]

It's been a few months since the last post. During this time, new AI tools focused especially on architecture emerged. lookx.ai and promeai were two that I tested. Sketchup also implemented a plugin directly into the program.
Now AI accurately recognizes hand sketches or lines generated by 3D software, or even depth maps.
The quality and precision of the idea we want to show is increasingly better and will evolve exponentially in the coming months.
It won't be long before AI can accurately place available finishes on the market, such as wood floors, fabrics, bricks and furniture that brands sell.
I imagine we are not far from having spaces designed entirely by AI, as there are also AIs that generate 3D files and floor plans.
Rendering for architects and interior designers will be the first to be affected, as they use renders to "illustrate an idea", (in fact, most show the idea only with a moodboard...) and most of the time they don't need to have an Oscar-worthy image.
For developers, I am still skeptical, as they require very precise images of the product being sold, with perfect humanization and even videos.
In the coming months we will be inundated with AI-generated images, like this @miladeshtiyaghi profile. Architects can benefit in some way, transforming stacked boxes into photorealistic images with just a few clicks.

[Rodolfo Tanno]

On 8/7/2023 at 11:50 AM, Francisco Penaloza said:

Talking about ownership, do you know that there is a dataset for stable diffusion that is called MIR. As you may think, yes is a dataset created using renderings from the studio MIR that has a very distinctive style. MIR website is very minimal, so I can't corroborate this information.  But if this wasn't created by MIR studio, can you imagine that in a very close future, anybody could create something that looks like your work, that is beyond traditional inspiration.

 

 

 

The MIR website is currently flooded with images made by AI.

[Mika Kerr]

I stumbled across this post while I was looking for some new tools actually!  

I felt like commenting since things are moving so quickly. Everyone makes great points, but I'm surprised at how accurate Scott's assessment is even a few months later with all these changes happening every day.

For me -  AI just helps me do my job faster, it doesn't take business away. I can iterate through ideas, create mood boards, and quickly show my clients designs at fraction of the time and cost it used to take. 

The tools HAVE gotten better though. A few of them can render sketches and basic 3D mockups surprisingly well. They are also crazy photorealistic and the telltale signs of AI are less easy to spot like Midjourney (if you know how to work with it), Deft (for exterior and interior generations), or just running a powerful instance of stable diffusion if you're tech savvy (too hard for me). I mean, look at these screenshots just from drawings and mockups. Crazy how accurate they came out.

The thing that's most helpful for me is when I render the look and feel of a home prior to doing a ton of work. Can't tell you how many times it's saved me hours and probably made my clients happier. 

I think there are lot of improvements that need to be made if it's going to do architecture or structural drawings more broadly, but for generating inspiration, I think it's here to stay. It is a valid point about copyrighting though....I'm curious how all of this will play out in the courts. These models are likely all trained on someone else's hard work. 

Edited February 8 by Mika Kerr