Implementation of generative AI that automates the production of architectural sketches, optimizing working time and freeing up resources for tasks with higher added value.
Context
In architectural tender competitions, creating sketches by hand is essential. However, these are repetitive, often non-creative and time-consuming tasks (3d/sketches) so that Egis limits creative iterations on these sketches. A firm can devote more than 274 days to making these sketches, which can represent up to 48% of the time of architects.
Egis seeks to automate this step to allow operational gains and to allow several versions of the sketch that engineers can iterate on.
Challenges
- The tool should be able to extract and interpret images from Google Street View and other CAD software.
- AI must generate sketches while respecting the architectural signature of Egis and architectural firms.
- Implement and master state-of-the-art models in image generation (Stable Diffusion algorithms).
Our approach
A multidisciplinary Theodo Data & AI team worked hand in hand with Egis:
.svg)
Right First Time
Automated transformation of 3D images and Google Street View views into Black & White sketches while respecting the artistic signature of designers.
.svg)
Just-in-Time
Using advanced models for segmentation (Segment Anything), edge detection, and edge generation (ControlNet, Stable Diffusion), Theodo has developed a solution that automatically converts images into sketches in a precise style.
.svg)
Learning Organization
The key advantage: the interface, based on Gradio, allows illustrators to quickly obtain sketches that can be used in their creative tools such as Photoshop.
“A fruitful collaboration with Theodo Data & AI and Egis that comes to fruition with the collaboration award”
