3D Warehouse & Gemini AI

Off Topic
,
1

Just had 10 minutes fun with a screen capture of one of my 3DWH models and Google Gemini AI with my Samsung S10 tablet while having a morning coffee.

So quick and easy.. really makes me wonder the need for any graphics software any more.. Adobe, Twinmotion, 5D, Vray, Affinity, Canvas etc must be all wondering their future.

Screenshot_20260416_120705_Google
Screenshot_20260416_120705_Google1846×1598 327 KB

Thought I’d try another practical example.. the image is from a site visit 10 years ago to a shophouse complex under construction in vietnam, the building shells are complete, the theme was to be “New Orleans.”

Screenshot_20260416_123931_Google
Screenshot_20260416_123931_Google1752×2800 456 KB

Here are the before and after for better comparison

Screenshot_20260416_122940_Photos
Screenshot_20260416_122940_Photos1920×1201 276 KB

1920×1201 575 KB

I am surprised at its choice of planting, I did not tell it it was in a tropical location

Thankfully, I am retired.. what would have cost thousands $ of effort and experience is undermined in 10 minutes. I don’t like my chances of justifying to a client a $30k facade and streetscape design fee for a 10 line prompt.

2

Well, if your daily job only consists in creating random pictures of cars having 6 bolts instead of 5 to hold their wheels in place AND/OR creating rendering of buildings in which windows suddenly turn into doors of non-existent balconies AND you don’t need any sort of control over the final output AND you only need very low resolution, THEN I guess you should replace any professional tool you’re currently using with Nano Banana ASAP.

3

Every project I’ve worked on has had at least ten views (front, back, bird’s eye, worm’s eye, and so on), and 95% of them have been revised at some point. Add a window, change a material, re-render the whole set. Multiple times.

A typical house project ends up with 20+ stills, a video comprising 20+ shots, and 20+ panoramas stitched into a walkthrough tour. Then the client requests revisions. Twice, at minimum.

That’s the reality of this work. So when I hear AI rendering praised as a workflow revolution, I struggle to see where it fits. My view, based on producing architectural stills, videos and panoramas since 1991, is that one-off AI generations are largely useless in any real-world production scenario. They look impressive for about five minutes, right up until you notice the kind of artefacts Panixia highlighted.

Will it improve? Certainly. The obvious failures will get ironed out over time. But until AI can produce multiple consistent views of exactly the same building and then reproduce that same building again with only the revised elements changed, it remains a hallucination generator with a convincing paint job.

4

Seems I touched a nerve ! I am not praising it, I am warning about how it threatens your job, if any client can in 10 minutes generate realistic impressions of their designs how are you going to justify $$$$ in doing the same task.. and expect clients to compare your work constantly to AI throughout the life of a project. At the monent this just considering the superficial aesthetics of a design and it is getting pretty good.. Soon it will be testing your skills at the spatial design, code compliance, materials selection, etc etc. As I noted I am retired.. thankfully..

I am well aware of the reality of work, leading hundreds of commercial projects from concept to completion..

5

Story if I miss took your comments but I stand by what I said, the nearest analogy I can think of right now is …

Now that sub 6 second 0-60 EV cars are available to everyone does that mean that Formula 1 drivers are at risk? Or does their consistency at knocking out blistering lap times every other weekend mean that irrespective of what we ‘think’ we can do they’ll always be better high speed drivers than us?

6

Interesting analogy, but there are billions of drivers and only 22 F1 drivers.

All I am saying, if your are a professional be prepared to be challenged in terms of cost and time by your clients resorting to AI answers to their concerns.. 10 years ago, the client could not generate that AI scene.. he had to use my skills to do that.. and it meant say 40hrs work and $$$$.. now he can bypass me, test 10 different alternatives with his inhouse team in a day and send the preferred option to me to document.. and I am confident it is happening right now somewhere in the world.

And I can assure you that some of my very wealthy former clients are probably dabbling around with AI right now…

But, yes there will always be exceptions to the rule and F1 drivers in every profession.

7

Well, I think you are missing the point,

Adobes share price has collapsed.. many contributing it to the rise of AI generated graphics doing what used to take expensive software and skilled professions lots of time and money to do!

In that simple example of the car AI did exactly what I asked of it, nothing more, nothing less.. in 10 minutes of effort and maybe 5 secs of processing time. If in the recent past I wanted to achieve that result,it would involve significant hours and software.

In the second example generated from low quality site visit snapshot it did exactly the same, respected the photo’s camera position, lighting, building form, fenestration and theming instructions.

That speed, accuracy of interpretation and ease of access is pretty significant to me and I guess other professionals also.. expecially how it might affect your business revenue, client services and role and status in the construction industry.

PS, I am well aware of AI’s limitations, I have posted examples on this forum before.

8

Probably AI will take the job of a lot of people, but I don’t think AI can replace a good architect, a good design implies more than intelligence, there are emotions and feelings that AI can not replace. I don’t think that AI can make even a brutalist design better than a good architect.

9

The problem is that AI is still too loose and fast with the products it generates. In a word, it is too chaotic and not deterministic enough. I’ve watched several videos now of AI trying to generate trusses in SketchUp for a roof. It can mostly get it right but it will give you something slightly different each time you duplicate the prompt.

This works for some things but not everything, it is especially annoying when you are really trying to nail down and fine tune something. The variability or fuzziness of AI is confounding in certain applications. Sometimes all you need is a “toaster” to get the job done. Imagine if MacOS, Windows or even SketchUp reacted unpredictably every time you used it or gave it the same inputs, expecting the same output.

In a sense this randomness or chaotic behavior is a sign of intelligence however if it becomes too chaotic it actually looses its usefulness. Higher intelligent organisms like ourselves are in many ways very similar. We do have some instinctual and hard coded behavior however we are able to use logic, reason and our personal and moral agency to overcome these hard coded tendencies. Conversely is we suffer from mental afflictions such as schizophrenia or other more extreme forms of mental illness we become chaotic, our thought processes lose coherence to the point that we are unable to function in the real world.

At some point AI will probably become intelligent enough that it can self reflect and realize when it needs to become more exact or when it can become fast and loose with its responses, and get away with it. When that day comes, yes, we will all lose our jobs.

10

you’re an advanced user making big project where it’s worth taking time to make proper renderings. Many active users on this forum are in a similar situation.
I train a lot of interior designer, people who do small jobs, 500 - 1000€, they need a quick way to make a 3d model, because time is money. and once they’ve done it, they need a quick way to show an image or two to the client so they can pull the trigger.

For them, Esncape is great, chat gpt is faster and does the job.
they only need one or two images, they need to go from start to finish in a couple of hours. they are ok doing some minor post processing on a couple details that didn’t go as planned.

I talked to some of them last week, they tend to use ai tools, if the client wants more images, they can make actual renderings for an extra cost. I get it, at their scale, their price range, I get it.

I agree, yet for some people, it’s more than enough. they’ll tell you, bette an hallucination in 30s than spending 3h on a rendering.
the same way people will order a big mac instead of spending an hour cooking I guess :slight_smile:

I don’t see Rendering software dying anytime soon, but their target might change a bit, with integrated AI tools for a quick result, and more advanced tools for bigger needs.

kinda like Veras in Enscape, Veras produces okay images without having to deal with a lot of parameters.

11

I’m with Nab here.

I recently completed a quick job for a clients gift shop. They wanted to explore the possibility of some new shelving. All I had were some rough dimensions and photos.

I modelled this, exported it as a JPG and sent it to them. Job done!

Realistic barn gift 0
Realistic barn gift 01620×1154 350 KB

They didn’t need a Render but out of interest I’ve just run it through CoPilot.

After 4 attempts (spanning about 10 minutes) it’s given me this.

Realistic barn gift 4
Realistic barn gift 41536×1024 447 KB

It’s not perfect, it’s messed the flooring up and the finishes aren’t correct. I could probably get Copilot to correct it if I continued with the process but I guarantee if I sent something like this to the client they’d be very happy.

12

Actually is the other way around. I said that IF your job is only creating absolutely generic, low resolution and uncontrolled whatever, THEN it will take your job, otherwise it will not (at least for now).
And that’s it.

No way. LOL

The camera perspective is different, the lighting is rotated about 90°, the building form was completely made-up from hallucynations and the fenestration doesn’t match at all.
So.. yeah.

13

Not even close.
An half-skilled user can take that EXACT shelving model and photomatch it properly in a matter of less than a couple of hours, including modeling, matching, rendering and postproduction. He doesn’t make the errors that the robot did here and can iterate consistently if reviews are required.
In addition to that, would have kept the camera straight to begin with.
But if you are fine with that for your needs, that’s absolutely fine.

14

It’s not my needs, it’s the customers.

If they are a small business ( a lot of my clients are) and just want a general idea of what their store/piece of furniture “could look like” then its up to them.

If they are happy with a screenshot of a Sketchup Model or a few PDF section views via layout then an AI generated visual will probably suffice.

15

It’s the same. If you need to fulfill the need of a customer, then it becomes your need.
Other people have different needs indeed.
In that case you cannot get away with AI slop (although you can maybe use AI for other things different from just prompting “Make this image photorealistic, make no mistakes”).

16

Yep, you are missing the point !

17

Well in the quick example AI certainly did a good job of taking a bare carcass of a construction site and applying a " New Orleans " themed aesthetic over it.. of course it did things that are not practical, code compliant but in 10 seconds it summarised the essense of the theme as requested in the prompt. Colour, variety, detail enough to form either a brief from the client, “thats the look I want to achieve” or a submission by the archiect to the client " is this style you want to achieve " both are valuable.. and achieved with minimal effort… and this will have fee and role implications for any architect.

Having an AI tool that produced these results would have dramatically increased the clients ability to refine and define the design direction and instructions (design brief) to the documentation architects.. depriving them of a significant part of their design development work.

PS, I asked it to do a brutalist version

1776480078503
17764800785031304×816 281 KB

18

Try doing a traditional Japanese (Kyoto) look to it, I’m curious how that would come out.

19

I wasn´t referring to any architect, I said that only the good ones will always be able to beat the AI. You can add all the prompts you want but a machine will never think or feel like a human.

You´ve asked to an AI to design facades based on an existent construction, ask it to do a functional and beautiful design out of nothing like we must do. Facades alone aren’t architecture.

20

I am not disagreeing with you, just highlighting a practical example use case and how that might impact architects., their fees and roles in the construction process. I have my own personal AI (the missus) looking over my shoulder while designing a country retreat for us, I expect increasing AI will feel similar .

Thankfully she does not follow the Sketchup forum :slight_smile: