It was pretty exciting when, 11 years ago, a new kind of CAD company emerged, several seemingly at the same time. They were pure cloud-CAD plays, taking advantage of new capabilities in Web browsers that allowed us to select and edit entities. The companies had names like sunglass.io, TinkerCAD, To3D, and Onshape. All ended up being acquired by larger firms.
After the Open Design Alliance cracked the RVT and RFA file formats of Revit, a second wave of CAD companies emerged two years ago, with names like Infurnia, Snaptrude, and Qonic. All three handle some aspect of architectural design through Web browsers, and all continue to develop their BIM-related software, having obtained investor funding.
Now a third wave is emerging. These upstarts are taking advantage of AI software like ChatGPT and Stable Diffusion to produce CAD programs that also run in Web browsers, and that read our written and/or sketched inputs to generate 3D rendered images. Three that I have come across are Leo, Maket, and CadAiCo.
Two of them, Leo and CadAiCo, work like this: you enter a prompt text and then the program either succeeds or fails at producing what you want.
- Prompt text is a brief description, such as “a box to hold cars.”
- Success means it produces something that agrees with your intentions.
- Failure means it doesn’t understand your request, or else produces something functionally useless or artistically improbable.
The other one, Maket, prompts you to select options from dropdown lists. Let’s take a look.
MCAD with AI: Leo
Leo (www.getleo.ai) calls itself the “world’s first engineering design copilot powered by generative AI.” It goes on to say, “If you can imagine it, I can design it, from parts to fully assembled products, with 3D CAD models you can edit anywhere.” It is in public Alpha 2 stage.
I tested it by signing up for a free account, and then entering text prompts, like this:
Leo: What would you like to design?
Me: A box that fits a car
Leo: I’m sorry, but I’m not quite sure what you mean by “a box that fits a car”. Could you please provide more details or context about the product? What is the purpose of the box? Is it for storage or for transportation?
Me: A box to store an automobile
Leo: Got it. Here’s the product description with some additional details:
Description generated by Leo from my brief prompt
Leo: If you’re satisfied with this product description, you can add design inputs below or just click the Generate button.
I clicked Generate. After about a minute, Leo produced images of shipping containers. None of the storage units were made of concrete. I don’t know the source of these images.
Three container-looking storage units for my automobile displayed by Leo
Elsewhere on the screen, Leo generated the specifications, of which I reproduce a portion here, I don’t know where Leo gets the data from.
Portion of the technical specifications for the storage unit
Next, I use the sketch input option to draw badly the shape of a VW-like automobile, and then clicked Generate again.
Leo includes a sketcher for describing our ideas visually
After a minute, Leo produced these funky container-looking vans:
Storage units after being modified by my sketch
Finally, I asked it to color the vans red and yellow:
Me: color it red and yellow
Here Leo regressed, losing many of the car-like attributes from the model, making it look more like a trailer:
Wheeled storage units changed from gray to red and yellow
Each image generated by Leo can be further refined though a set of actions that are kind of hidden until you pass a cursor over the image:
Toolbar of additional actions
The actions are Approve, Disapprove, Magic Eraser, Generate Variants, Download, and Share. The Magic Eraser option let me use a marker-like tool to white-out portions of the image. When I tried, it didn’t seem to make a difference to further modifying the models; I suspect this function is still being worked on. Generate Variants makes more changes, such as the set illustrated below, which spaces the tires further apart:
Variants generated from my selected model
Download didn’t seem to work for me. Share sends the selected image to social media; a Copy to Clipboard option lets me paste the link to the image elsewhere.
The Gallery conveniently keeps copies of all the images I’ve generated.
The promise on the Web site’s home page of “3D CAD models you can edit anywhere” did not seem to be operational yet; at least, I couldn’t figure out how to access 3D model files to edit elsewhere. As Leo is in Alpha 2 stage, I am guessing this function is yet to come.
AEC CAD with AI: Maket
Maket (www.maket.ai) generates floorplans, 3D models, and renderings from existing drawings of buildings or else by starting from scratch. It allows five projects for free, but the free plan excludes things like DXF export and AI processing, and limits the quality of renderings. For unlimited projects and functions, the cost US$288 a year — or $30 one month at a time.
Initial starting points for Makit
The very first step in creating a floor plan is to specify “constraints” (its term), which I would call design parameters, like the number of floors. I accepted all the defaults, and then clicked Generate. Makit generated two floor plans that, to me, looked identical:
Two floorplans generated fromMakit’s default design parameters
I picked one, and then chose styles for an exterior rendering: Scandinavian style, cedar cladding, located on the coast, and displayed during the day. Maket generated four renderings for me.
I picked the one I liked, even though it lacked a coast (see figure below), and then found there were several more options available: Edit brings up a solid brush tool to draw fat lines on the image and nothing more; I suspect this will be improved in the future. Reimage generates three more variations. Report isn’t working yet. Export saves the image as a JPEG, and not, unfortunately, as a lossless image like PNG.
Additional options in Maket
At this point, I could not figure out how to go further, even though the progress meter said I was at step 2 of 5.
I had wanted to test the AI function that adds dimensions to existing floorplans, but found that Makit (in this iteration) imports only image files. After I converted a CAD floorplan to PNG and tried importing it, I found I needed to pay $30 to proceed further.
MCAD with AI and Blockchain
CadAiCo (www.cadai.co) is so new that it is not yet operational; the Web site just tells us what it plans to do. So I couldn’t test it out. What the company hopes to ship is an AI-based CAD system that combines machine learning with blockchains. It plans to operate according to the figure shown below.
How CadAiCo proposes to integrate AI, CAD, and tokens
CadAiCo says its AI system was trained on millions of CAD models, although I don’t know the source of them. Further, the company wants to enable payments among customers and third-party developers through the $CAD token, but $CAD doesn’t seem to be listed yet at coinmarketcap.com/new.
It also plans to produce APIs that integrate into today’s desktop MCAD systems; in the figure below, note the use of the word “will.”
At this point, CadAiCo is a promise.
PS
ReRender AI is another one I tested (rerenderai.com), but found it has no hard link to CAD, despite the halo image on the Web site’s front page. It took an image file that I uploaded (at top, below), and then applied a theme to it (result at bottom).
Before and after ReRender AI applying a theme to one of my photographs
As there was no ability to import CAD drawings, I did not test ReRender any further.