Text to 3D CAD is coming: Who's building it?

A look into who is creating a new way to build CAD models

Engineers and hobbyists have been interacting with CAD models by pointing and clicking for decades. With the power of AI and machine learning, we’re now seeing platforms arise that allow users to enter a prompt and have a CAD model appear. If you’re experienced in CAD, or have even just taken an introductory CAD course, you can probably see how this technology could: 1. Increase the speed that CAD models are developed, and 2. Lower the barrier to entry to creating CAD models.

So who’s building this?

1. Zoo

Zoo (formerly known as KittyCAD) is a startup focused on creating ‘infrastructure for hardware design’. Led by CEO Jessie Frazelle and cofounder Jordan Noone, Zoo hosts a variety of browser-based CAD applications, and they’ve just launched their text-to-CAD model. Let’s try it out:

A simple CAD model I generated with a prompt on Zoo

With a simple prompt, I’m able to generate a CAD model within seconds, and it’s ready to download in a variety of formats. I’m able to rate my design with a thumbs up or down, which I assume is used for future ML training (Note to anyone at Zoo who sees this: Why is thumb down on the left?). I’m not able to further edit the model or adjust parameters within Zoo, but it sounds like these features should be coming in early 2024.

Zoo’s text-to-CAD can’t handle too complex of requests yet, but I expect this team to make significant progress this year. Their website is also incredibly cool (I love the font choice) and it all around just feels like a platform from the future. Zoo’s text-to-CAD is free at the time of writing, but is switching to a free trial and paid API on 01/15/2024.

Mancer Labs

Mancer Labs is a bootstrapped startup founded by Adam Hibble. Adam has demoed a text-to-CAD model on his Twitter (are people calling it ‘X’ yet?) that takes a text prompt and produces a CAD model. Not only is the model downloadable, but the demo boasts the ability to edit the parameters in real time within the browser. This demo is exactly where I see the future of CAD modeling heading, with AI producing an impressive boilerplate, and a human editing that boilerplate to meet the needs of a project.

A screenshot from Adam’s demo on Twitter/X (@Algomancer)

Adam’s model is not yet publicly available, citing compute as a current bottleneck. When he is able to bring it online, he’s stated he plans for it to be open source 😯. This seems to follow a trend we’re seeing in the AI industry as a whole, where closed source and highly funded companies (e.g. ironically: OpenAI) currently have an edge over open source and less funded startups.

OpenOrion

OpenOrion is a 1-man startup comprised of Afshawn Lotfi. Afshawn is working on parametric auto-differentiated CAD (No lie, I would not have known what that means if it weren’t for his recent post on Twitter/X). Afshawn is one of my favorite follows on Twitter, because I feel like I’m watching a genius at work as he shares progress and insights into what he’s building. Something he has already built is a text-to-CAD model called CQAsk. CQAsk is open source and is powered by OpenAI’s LLM models. Along with the prompt to the LLM, is also takes input for CAD specifications.

An example prompt and model from CQAsk

I’m not sure that Afshawn is further progressing this model, but I’m sure text-to-CAD will certainly be featured in the big project he’s cooking up.

Future players?

I think that the power of text-to-CAD is undeniable and that we’ll soon see more companies innovating in this space. I think that Autodesk will undoubtedly incorporate text-to-CAD into their programs eventually, but I expect them to be slower than their startup competitors. Since I’ve spent most of my 3D modeling time in OnShape, I hope to see them take a crack at this as well.

Final Thoughts

I think the text-to-CAD space is going to be an exciting one to watch in 2024. Text-to-CAD is certainly in it’s infancy stage, and I’m excited to look back at this post at the end of 2024 and compare to the progress that has been made.

As someone who spends 40 hours a week in AutoCAD / Civil3D, I’m excited (and maybe a bit nervous) to see who brings about these type of models for 2D CAD and specifically civil infrastructure design. Maybe I’ll build it myself…

If you made it this far…

Thanks for reading the first installment of The Built Future newsletter. I hope this is the first of many posts as we see more and more innovations coming to AEC technology.

Any questions, comments, corrections? Be sure to leave a comment below!