Revit AI Tutorial: Image to Model with ChatGPT 6 Astra

OpenAI just released a new model called ChatGPT 6 Astra. They claim it is AGI.

Users are having great results with Astra alongside open-source modeling tools such as Blender. How does it fare with Revit? That's what we're about to test.

Watch the video:

Setup

First, download the ChatGPT desktop app here:

Ensure you have a paid account, else you you won't have access to GPT-6.

Then, install a Revit MCP server. You can use the pyRevit MCP or a C# based one, such as BIMwright.

Test 1 - Apartment Layout

In the first test, we ask ChatGPT to create an apartment layout from scratch. This is based on a real project called Franky by Quinzhee Architects.

The first attempt is bad, with the fridge blocking the path.

After increasing the effort level and giving a few extra instructions, the second layout is better, although not as good as the real layout.

Test 2 - Cozy Cabin Revit model from Image

In the second test, we provide a few images of a cozy modern cabin called Repère Boréal, located near Quebec City.

This one is quite impressive: the model is very accurate. It even managed to model the furniture and elements around the cabin itself: chairs, the fireplace, shelves, plates, and more.

However, there is a caveat: it modeled everything as DirectShape, a type of model that cannot be edited. This is typically used in Revit for imported geometry.

So the modeling was accurate and impressive, but mostly useless as a model with the DirectShape technique used. In coming posts, we'll explore how well it does when sticking to loadable and system families.

Verdict: Is this AGI?

Probably not. And what is even the definition of AGI? What are the criteria?

Regardless, this is impressive. You don't need a huge imagination to see that the world of modeling will become more and more automated in the coming years.

What if you could learn Claude for BIM with 6 brilliant experts in the space? Join our upcoming Claude for BIM live course.

Learn more