AI command interface for Vectorworks

[Christiaan]

AI rendering is an obvious feature to add to Vectorworks, but what about adding AI to the interface to assist with modelling?

 

e.g. prompt: change plasterboard thickness of all Wall Styles in file to 15 mm 

[Matt Overton]

1 hour ago, Christiaan said:

AI rendering is an obvious feature to add to Vectorworks, but what about adding AI to the interface to assist with modelling?

 

e.g. prompt: change plasterboard thickness of all Wall Styles in file to 15 mm 

That shouldn't need AI just features like Materials to do the job they are said to do. 😉 

Save the AI for Fun things. 

[Leon You]

Great idea! @Christiaan 

Can we create something like https://www.blueprints-ai.com/ to "translate" blueprints to 3d models and vice-versa?

 

[Jeff Prince]

45 minutes ago, Leon You said:

Great idea! @Christiaan 

Can we create something like https://www.blueprints-ai.com/ to "translate" blueprints to 3d models and vice-versa?

 

 

You should check out the architect with the quoted testimonial's website LOL.

[BartH]

On 2/29/2024 at 5:26 PM, Jeff Prince said:

 

You should check out the architect with the quoted testimonial's website LOL.

I assumed it was an AI generated testimonial. Might be an AI generated company too.

It says a lot about AI's readiness for real work. AI is likely to be a "game changer" soon. But, remember until it works its way through the courts, we could still be held liable for using AI work professionally, without copyright, and without the consent of the people who unwittingly contributed to the AI model's creation. 

 

I enjoy keeping up with what generative AI can do. I hope to use it professionally one day. 

The OP's idea of using prompts for VW commands is intriguing.  Who loves the Auto CAD command line interface? 

 It would be interesting to test how "change plasterboard thickness of all Wall Styles in file to 15 mm" would work, sounds like real-time, Vectorscript.

I'd rather have a usable texture mapping tool, or 100 other fixes before VW starts committing millions of hours of LLM training data to an AI model. 

 

Bart

 

[VIRTUALENVIRONS]

Ai for design has been experimented with for over twenty years.  It has a value for concept design where the designer may have brilliant ideas, but not have a grasp of 3D design.  Even if it was perfected to react to human speech, I think advanced users would go back to the keyboard and mouse.

 

Artificial intelligence might have its own opinion.

 

Below is an example.

 

 

 

[Christiaan]

On 3/1/2024 at 12:35 PM, Matt Overton said:

That shouldn't need AI just features like Materials to do the job they are said to do. 😉 

Save the AI for Fun things. 

Except that you would have had to have set up Materials in the first place. AI could do also do that. I want AI to do the tedious stuff, not the fun stuff!

[Christiaan]

MVRDV are pushing the AI envelope. They mention the idea of integrating AI into BIM (at 15:50 m)

 

 

[Christiaan]

23 hours ago, BartHays said:

I'd rather have a usable texture mapping tool, or 100 other fixes before VW starts committing millions of hours of LLM training data to an AI model. 

I suspect this is like asking Henry Ford for a faster horse. AI integration could probably make a texture mapping tool far more usable than you currently imagine.

[Jeff Prince]

1 hour ago, Christiaan said:

I suspect this is like asking Henry Ford for a faster horse. AI integration could probably make a texture mapping tool far more usable than you currently imagine.

If that’s the case, I wonder which company is going to successfully develop the successor to the architectural design and documentation process.  AEC AI will probably follow the Revit -> Autodesk path, with a big fish swallowing up a smaller one.

 

Innovation like this rarely occurs in house, especially when radically different tech is being developed.  The old guard typically resists this and it takes outside eyes to shake it up.  The company with the deepest pockets is probably going to win.

[Matt Overton]

44 minutes ago, Jeff Prince said:

If that’s the case, I wonder which company is going to successfully develop the successor to the architectural design and documentation process.  AEC AI will probably follow the Revit -> Autodesk path, with a big fish swallowing up a smaller one.

 

Innovation like this rarely occurs in house, especially when radically different tech is being developed.  The old guard typically resists this and it takes outside eyes to shake it up.  The company with the deepest pockets is probably going to win.

Add the potential VisionPro is showing and you have the specs sheet for the last CAD program I'll invest in, in my life time. 

[Matt Overton]

2 hours ago, Christiaan said:

Except that you would have had to have set up Materials in the first place. AI could do also do that. I want AI to do the tedious stuff, not the fun stuff!

Or the Program ship from your local distributor with most of the common ones in your market.

Then you'd only need to make a small handful of uncommon ones each year.

[Jeff Prince]

33 minutes ago, Matt Overton said:

I want AI to do the tedious stuff, not the fun stuff!

 

Unfortunately, AI will be doing almost all the stuff within your lifetime.

Licensed pros will be required to sign off for liability purposes, until those laws change.

So, the tedious process of checking the AI "design" so stairs don't look like a rip off of a MC Escher design will still be needed 🙂

 

It's going to be like the transition from manual drafting to CAD to BIM...

only instead of outsourcing to India, you will be outsourcing to the cloud.

[BartH]

2 hours ago, Christiaan said:

I suspect this is like asking Henry Ford for a faster horse. AI integration could probably make a texture mapping tool far more usable than you currently imagine.

I don't currently "imagine" a better texture mapping tool, they simply exist, we just don't have them in VW. AI or not. 

( and on that 'Faster Horse" quote, it never happened. )
 

Quote

However, even if Ford didn’t verbalize his thoughts on customers’ ostensible inability to communicate their unmet needs for innovative products — history indicates that Henry Ford most certainly did think along those lines — and his tone-deafness to customers’ needs (explicit or implicit), had a very costly and negative impact on the Ford Motor Company’s investors, employees, and customers.

So, contrary to your point, Vectorworks does listen to user requests for innovation, that is the whole point of the "Wishlist-Content and Feature Requests" forum.

 

I believe that AI Integration into VW can't/won't make a texture mapping tool for VW. Engineers and coders would do that, with user testing and many iterations.

AI might give those engineers ideas on how to craft the user interface or implement code snippets from other sources.
AI might lay on top of existing code to make it respond to natural language rather than a mouse click. Fine, That works for me.

AI integration is simply another tool, of many, not a solution In itself. 

 

The AI-integrated BIM mentioned in the video above is exactly what I mean, AI didn't create the BIM software or its commands, one might use an AI LLM to convert natural language into a series of existing commands. With that said, AI is getting better at creating code. So, there may be a future where a generative AI model can, in real-time, convert a natural language prompt into an instruction set for another AI model to create self-destructive, Vectorscript objects, that self-execute, and destroy themselves (so the code base doesn't grow with every command). 
 

To @Jeff Prince's point, I don't think this can or should be the focus of Nemetschek/VW. We need them to focus on the core product. Let innovation happen in the fringes with more nimble creatives. 
I'm not quite as dire as Jeff in his prediction for the future. I think it will be more like graphic design when bootleg copies of Photoshop became available, suddenly everyone was a graphic designer, and almost all of them were terrible. I think there will still be a market for designers/architects who have the skill and talents to compete with the masses of AI-assisted creatives, but it will be harder. We will all lean on AI in some ways, but old-school raw talent and well-honed skills will still have value in the sloppy, Instagram-ready world of Generative AI. 
 

 

 

[Leon You]

In the AI era, the only things left for architects are ideas approved by the project owner and the responsibility for signing AI-generated products.

But what can we do now in this transition period to ensure we still have Vectorworks and our jobs in the more and more intense competition?

[Matt Overton]

33 minutes ago, Jeff Prince said:

 

Unfortunately, AI will be doing almost all the stuff within your lifetime.

Licensed pros will be required to sign off for liability purposes, until those laws change.

So, the tedious process of checking the AI "design" so stairs don't look like a rip off of a MC Escher design will still be needed 🙂

 

It's going to be like the transition from manual drafting to CAD to BIM...

only instead of outsourcing to India, you will be outsourcing to the cloud.

My first lecture in my first year of architecture started with that very line. 

30 years later it is only marginally more true today than then and fails to understand how much further it has to.

So sure maybe in my lifetime I might read about it happening while in the nursing home. 

[E|FA]

My first lecture started with "if you think you might want to do something else, try that first."  I didn't. 

[Jeff Prince]

3 hours ago, Matt Overton said:

My first lecture in my first year of architecture started with that very line. 

30 years later it is only marginally more true today than then and fails to understand how much further it has to.

So sure maybe in my lifetime I might read about it happening while in the nursing home. 

The number of high skill employees required to create and document a project has radically declined over the past 50 years, and that is largely due to drafting improvements.  When that part of the profession is further automated, fewer people are needed and the important “wax on, wax off” tasks that help train young professionals will be gone.  When people start letting AI do the creative part, they are slitting their own throats.

 

If you want to see the future, just look at manufacturing.  AEC has always lagged a good 30+ years behind that industry.

 

@BartHays mention of the graphic designers is another good example.  That profession has been largely converted to a commodity traded on price.  That’s been happening to commercial architecture for quite some time.

 

At any rate, it will be interesting to see how it all plays out and identifying opportunities to profit during the disruption.

 

 

[Christiaan]

5 hours ago, BartHays said:

The AI-integrated BIM mentioned in the video above is exactly what I mean, AI didn't create the BIM software or its commands, one might use an AI LLM to convert natural language into a series of existing commands.

Fair point.

[twk]

1 hour ago, Jeff Prince said:

identifying opportunities to profit during the disruption

couldn't have said it better..

[Matt Overton]

18 hours ago, Christiaan said:

Fair point.

In theory, the AI could already author a vectorscript description of the Wall type you'd like to replace the wall type you had if there was enough of the function exposed in the VS documentation. 

[BartH]

2 hours ago, Matt Overton said:

In theory, the AI could already author a vectorscript description of the Wall type you'd like to replace the wall type you had if there was enough of the function exposed in the VS documentation. 

There have been some threads on this already,

As I understand it, there isn't enough of a repository of Vectorscripts to train an AI model. 

AI is great for creating code in generic Python(etc) because there are millions(?) of examples available for scaping on the web.

There are relatively few that use the Vectorworks specific calls. So, the best AI can do is make code that "looks" like Vectorscript but is actually just junk. 

 

[geo44rey]

"...one might use an AI LLM to convert natural language into a series of existing commands."

 

Yes, that is exactly the point - really "AI" in that case is not really more than a language interpreter. And that is pretty easy (for me to say at least:), and pretty powerful.

 

One of the biggest beefs I have with the VW UI is how it suffers from too many "cooks in the kitchen" over a long time and often is inelegant. So, we get "dumb" things like

 

a) Too many lists of object types in something like File > Import that is not smart enough to figure out what kind of file one wants VW to import (or not)...

or

b) The Modify > Convert (or Compose or Group or some combination of all of those) menu, with 20 different ways to "connect those lines (or whatever) into something I can use", without one having to know what type of lines they are, why they won't connect, (does one really need to know if they are nurbs or polylines, grouped, etc, etc??)

or

c) One is working in 3D, with a foreground object highlighted, and when one goes to move that object, the UI pushes the selection to some other object way hidden behind.

(I mean you are working in 3D, and that thing that is selected and drawn like in real life is probably what you want to be manipulating, yes?)

 

An AI "interpreter" would be a decent first step to at least potentially globally and flexibly clean up each user's personal pain points with the UI.

 

AI tool interfaces like "make a straight stair from the 1st floor to the 2nd floor with maximum risers and minimum treads (please)" would not probably be a hard lift soon thereafter.