top of page

Introducing Generate: Creating Plans with Machine Learning

As of today, we start to roll out Generate to Enterprise customers.



What?

Generate is a new product from PlanFinder that leverages the power of Machine Learning to generate floor plan options. Instead of having to spend hours manually drawing a plan, multiple plan options are generated within seconds.


The process works in three simple steps:

  1. Assign the geometry of a floor plan

  2. Within seconds, you'll see up to 8 generated floor plans

  3. If you're happy, add them to Revit or Rhino

Architects can adjust the desired country and room counts to gain more control over the generated plans.


The Generate command adds to the existing Fit and Furnish commands. With Fit, architects can fit plans out of a database to their new designs. With Furnish, plans can be automatically furnished. With the addition of Generate, users can now create tailor-fit plan options to a unique input.



How?

To generate floor plans, PlanFinder uses Machine Learning, a field in Artificial Intelligence that lets computers learn from data. This is required, because plans look different from country to country, making any non-data-driven approach practically infeasible.


The algorithm is loosely inspired by Denoising Diffusion Probabilistic Models, a technique that lets a computer learn to gradually denoise an image. Models like these have shown to be successful at image generation, as shown by models like Dall·E from OpenAI.


Below here is a visualization of the Machine Learning model creating multiple floor plans for various inputs:




Why?

At PlanFinder, we believe that Machine Learning can enable architects to speed up their design workflows and increase creativity. Generate lets an architect produce multiple floor plan options in seconds, giving a quick insight into the space of possibilities, saving time to design better and more buildings.


If you're interested to learn more, don't hesitate to contact us at contact@planfinder.xyz, we'd love to talk!



899 views0 comments
bottom of page