I work in construction management and I think you’re underestimating the complexity of generating a set of construction plans that meets code, passes inspection, and has coherent aesthetics.

There isn’t just one set of building codes for every jurisdiction, different jurisdictions adopt various sets of code.

Different geographic regions require various things that other areas don’t require. My state doesn’t have earthquakes or hurricanes, but we do have to have stronger roofs for handling snow load. Buildings in Florida need specific methods to handle hurricane force winds. Buildings in California need specific methods to handle earthquakes. And so on. How a building is designed is highly dependent on where it is located geographically.

You’re also underestimating just how many different materials/fixtures/fittings get installed in a house. Plumbing fixtures and light fixtures, electrical wiring devices, floor/wall/ceiling finishes, doors and door hardware, siding (type, color, trim color), windows, woodwork, cabinet, cabinet hardware, countertops, bathroom vanities, appliances, rain gutters, garage door, driveway/sidewalk material and color, deck material and color, etc.

I guess what I’m trying to say is that designing and building a building is far more complex than it seems.

> underestimating just how many different materials/fixtures/fittings get installed in a house

This is one of the reasons why I would believe Lowes or Home Depot or another construction supply place would 52 such a piece of software. They have those zillions of "something" that get installed, and the lumber. They already know what all those people buy.

At least for a construction company: pick a general house, gives you adaptations for common landscapes, choose interior finishing(s) / style(s), gives you plans (with adaptations), already has them approved for your area (with variants), gives you parts list, gives you plumbing/wiring, gives you button to put it all in your "cart," drive to store, pick up house worth of stuff, Lowes/Home Depot make a quick $50k-100k or so.

If you build a development, you say, give me 10 of that. Build it again, say I'd like my old order.

what does 52 mean when used as a verb, like in your comment?

Visually, its supposed to be a heart shape. Such as "I (heart/want/like/upvote) this" What I would describe as mild 1337 speak, using limited ascii art when it shorthand's an idea. "I 52 this idea."

Depending on how much you like symbology, since it's 52, it also carries playing card connotations, with "makes a good hand", "hearts the card game", "want this to shoot the moon."

> my state doesn’t have earthquakes or hurricanes, but we do have to have stronger roofs for handling snow load. Buildings in Florida need specific methods to handle hurricane force winds. Buildings in California need specific methods to handle earthquakes.

If you tried to accomodate all of these methods now, your looking at much higher costs. But if your open source plans include all these methods, and people can produce kits that are shipped to you (either raw, or partially assembled) because they are identical, the costs would quickly start dropping..

There also seems to be shortage of cad tools avable to inspect a building plan.,like as a consumer you would home you would get a plan of the house when you buy the house

CAD is used to create construction drawings, but once they’re created, everyone works off of a PDF set of plans. You only need the CAD files and software if you want to change the plan yourself.

It’s also fair to say the training data is probably not readily available either

Seems like most local planning offices would have a lot of training data, possible even publicly accessible via foia

> I work in construction management and I think you’re underestimating the complexity of generating a set of construction plans that meets code, passes inspection, and has coherent aesthetics.

And yet we were able to make Figma.

Yes, we can achieve simpler tasks. This one is more complex.

Why is that surprising? It's like why AI art is a running joke to real artists, and LLMs are a joke to good software engineers etc.

Random text/image generators have no intelligence, no knowledge of design, building regulations, engineering, physics. A fun little tool to set up boilerplate is its peak usefulness.

There's a specific disease where it is incredibly easy to think that other fields you aren't an expert in are trivial.

If this is aimed at me, its not that I believe the act of creating generative architecture software is trivial. Perhaps too flippant above. I believe it is probably quite challenging. Which is an excellent reason to create such a software, because it's much harder for someone to quickly steal or release a competing product. I was only surprised someone was not immediately using this thread to advertise their house generator.

“Quite challenging” is an understatement. The problem is far too big for the tech startup model. You would need to get a bunch of expensive skilled engineers, and a bunch of expensive domain experts (from many domains, since “building houses” is a massive problem space with thousands of sub-domains), and work on it for probably years before you had an MVP. It’s not the type of startup that easily attracts investment.

Even if you could find investors,I'm not sure you'd find any customers.

Firstly, only a tiny fraction of people are interested in building a house. And the number that build 2 are a rounding error.

Developers already have architects on staff, already have libraries of plans, why would they pay for this?

Then, every detail of the generated plan would need to be checked. Every. Single. Detail.

Most people who build a house do do because they want to make a mark, or they need something they can't find. They are all literally edge cases. They'll sit with an architect for hours trying to get that dream out of their heads onto paper, adapting to limitations of budget, planning approval, local regulations, budget, site, budget and so on.

There is no market for a product like this.

That's been the Holy Grail since I was a kid (AutoCAD jock, mid '80s). I think the best we'll achieve is some themed parametric models. For stuff like massing, space planning, budgeting, and presentations.

Or an app with some dials and checkboxes for a constrained design space. Like for micro homes. Or vertically integrated companies like Lindal Cedar Homes. Kind of like buying a customized airplane or RV.

But for general purpose construction documents? I'd bet no. So many different construction codes. Site planning. Construction tech and products are constantly changing. Customers are psychotic. Etc, etc.

Disclaimer: I was just a drafter working misc A/E/C jobs. And I wrote add-ons for arch and civil engr. But mosdef am not an architect. Would like to be proven wrong.

Probably commercial BIM plugins for something like AutoCAD are doing stuff like this, although it's probably more like classic constraint solving than stable diffusion. Anyway BIM has been around for years, so that's the training set you'd want for doing other things.

I built a house (or more correctly I paid a company to build one for me).

The initial design could be done using an interactive tool that you can use. This is not different from web tools used to design a kitchen. I also think IKEA uses one.

I live in Scandinavia so it is probably different from what you know.

The company designing the house took our drawings and ideas and created drawings and an excel sheet they used to calculate the price of the project. A tool to do this would be valuable and same the customer some time ad I would be able to do most of the work designing the layout.

After the contract is signed the company would make proper plans used for building the house etc..

The complexity of the whole project is enormous but the initial planning would be a good fit for an interactive tool.