Constructive Intelligence: AI Floorplans 2D - 3D Current State of Tech
Sarah Murray
Head of AI Product @ REA Group | Technology, AI & Product | Speaker on Design & Technology | Investor & Board Advisor
The transformative capabilities of artificial intelligence (AI) in architecture and real estate are gaining momentum, specifically through the use of machine learning (ML) and large language models (LLMs) to convert 2D floorplans into dynamic 3D models. The convergence of these technologies enables more efficient and engaging design processes and property visualisations. It is important to have 3D representations because they help bridge the imagination gap.
Having spent many hours as an architecture student myself, painstakingly creating 2D plans to have then start another process to create 3D (while this is improved with tools like Revit at the time of design), If you want to go from 2D schematic designs, the typical floor plan designs you might see on a property listing, to 3D there is often a lot of manual work involved. There are also two different fidelities: photorealism, which is used for marketing and can take time to render, and 3D for experimental representation, which often requires expertise in additional tools.
In this article, I will shed some light on how AI, both traditional Machine Learning and Multi-Model (can ingest images and text) Foundation Models, can save time and money and inject some joy into creating these important assets.
Please share tools you have seen and work with.
How AI Transforms 2D Floorplans to 3D Models
Machine Learning (ML) Techniques
The use of machine learning for converting 2D floorplans into 3D models is grounded in specialized algorithms that have been trained to understand spatial relationships, architectural semantics, and design patterns. Here’s how ML functions in this context:
Example in Use: Autodesk uses these ML principles in tools like Revit, which helps architects automate tedious tasks like generating structural components from a 2D schematic, saving significant time in the design process.
Large Language Models (LLMs) in Architectural Design
While LLMs like OpenAI’s GPT or Google Gemini models are not traditionally associated with spatial modelling, they are finding unique applications in the design and construction industry. Here’s how LLMs are making an impact:
Example in Use: Higharc is exploring how LLMs can work in tandem with ML models to automate design compliance checks and generate design variations based on user input, which is beneficial in the early planning stages. (Beta-stages)
领英推荐
Comparison: ML vs. LLMs in 2D to 3D Conversion
Integrating ML and LLMs for Enhanced Design Processes
Combining ML and LLM capabilities holds immense promise for the architecture and real estate sectors:
Example Collaboration: PLACE Technologies is using LLMs in their Digital Housing Developer Archer to not only generate 3D models from 2D but also provide text-based interaction to allow the purchaser of an MMC development to customise the model with text commands.
Benefits of AI-Driven 3D Modeling
References and Further Reading
Generative AI Solutions for real-world problems.
4 个月About 4 years ago, in the pandemic, and after looking at the first Dalle (community v1) text to layout images I made I started thinking: pixels are the worst representation for design. A few months later I had this: https://arxiv.org/abs/2303.07519 And this was with neo-125 and gtp-j. We now have Claude and gpt4o. It's wild to me designers still haven't picked up on this, it's right there for the taking and very straightforward to implement.
Computational Design Lead at i2C Architects
4 个月I’m intrigued by the idea that Autodesk integrates machine learning principles into tools like Revit. Could you share any examples or evidence of this? I’d also love to know if the same is true for High Arc.