AI and Chief Architect Renderings


wesdutka
 Share

Recommended Posts

Before I modified the images above I did a quick test on Nano Banana to add some specific landscape in front of my crap plant placements with CA 2d plant images. I did not save that out since it was a quick test but it was very impressive what it did on the first pass/prompt.

Link to comment
Share on other sites

  • 1 year later...

Just thought I'd show some examples of a chief rendering and the results from ChatGPT and Gemini. Seems to me they're all currently producing quite similar quality. Also asked for some design changes like the fridge and oven cabinet on the left of the long kitchen counter and the down-lights inserted into the bulkhead above the counter. It took a few attempts to get it to do what I was asking for and some prompts I had to change completely so they were interpreted correctly.

Chief

image.thumb.png.63b8cb0dd8805e03fffa6aaa7193c351.png

 

Gemini

image.thumb.png.8689cae509e3f405097e3081e4837702.png

 

ChatGPT

image.thumb.png.53618372d968bdb6b23daea4ad6faa13.png

 

Chief

image.thumb.png.df9f36f42e2c5de2cc5afd8502d8bb76.png

 

Gemini

image.thumb.png.61fe6c683b50b98f76b3fe407b591291.png

 

 

Link to comment
Share on other sites

Here is a project that I am working on. I decided to try and get this closer to a selling point without creating a bunch of custom elements. This is early design development so I want to be time efficient in testing out the idea with my client. Here is the prompt and the 3 images that made the client call me back right away elated. Less (work) is more.

 

The first image I will load is an example of the style of house the client is looking for. The second image is a new design for the project. Please render the new design image with more of the character of the example image. Do not change the proportions of the new design or any of the major elements like doors, windows or roof shape. The roof eave overhang can be adjusted to bring in the additional millwork details and curvature of the example image. Add more detail to the front porch to make it more similar to the example image. Keep a loose architectural presentation style using the rendering style of D.K. Ching and similar architectural style elements.

20261016_155332.jpg

persp base 1 10 29 26.jpg

Carpenter Gothic cottage architectural study.png

Link to comment
Share on other sites

Interesting article Alan... and thanks for sharing the prompts.
 
...as for being cooked, it's all perspective. Yes, graphics will improve etc... and everyday people will get a hold of the technology and take a stab at designing on their own, but a good designer can never be replaced.
 
Give someone a stradovraious and say play it... and guaranteed they wont ever know what or how to do with it because it isn't thier skill set compared to a life long master.
 
At the end of the day, AI is just going to be another tool in the box. A computer model can simulate geometry and code all day, but it doesn't possess the real-world intuition, or years of field experience that a client actually pays a professional for.
 
Those who know how to build, detail, and actually problem-solve in the real world aren't going anywhere, and besides, an AI can't hold a professional license to stamp the drawings anyway.
 
That said, everyone should keep honing their skills cause this vantage point is truly affording us all leaps and bounds in terms of design presentations. Back to the thread, Gemini and Chat seem to be running a neck and neck race on rendering quality but it all depends on the guardrails and directions.


 

  • Upvote 1
Link to comment
Share on other sites

On 10/31/2026 at 8:29 AM, Gawdzira said:

Nice try, but 100 million homes is minuscule when you’re talking about training a model to understand architectural design. It would need to learn how and why homes vary, how those choices affect usability, and how the floor plan connects to a functional, attractive roof design, room sizes, circulation, structure, and natural light. There’s an enormous number of ways those pieces can interact, and a collection of finished homes doesn’t necessarily explain the reasoning behind them. We simply don’t have enough usable data for a model to understand that whole design space well enough to take over our industry.
They need billions of homes as well as a large data subset explaining the correlation between floor plans and their functional and aethetic aspects. 
We are a LONG ways off
They weren't the first to think, hey what if we trained a model on every house plan out there. Surprised it got funded, fairly simple for us to see why its not working...unless you want a bardominium, which is an easy task for Ai
As it to redesign a wing of a home to make it a 3 bed 2 bath with a laundry room and you'll laugh at what it comes up with and then cry when you consider how a roof works around it

 

Edited by Renerabbitt
  • Like 1
Link to comment
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
 Share