Using VQGAN+CLIP to generate axonometric drawings

On the contrary to the previous generation of wide-spread creative tools, such startling advances and research are occurring in artificial intelligence and machine learning algorithms that the way creative people, be it designers, architects or filmmakers, have interacted with these tools have been increasingly less monolectic. Instead of dictating the software what to create, these new dialectic tools enable creating the unimaginable, the unexpected, the surprising.


It is this playful type of interaction that will push creativity further, distill immense amounts of data into production. In this research, we will use one of AI algorithms, namely VQGAN+CLIP, to generate axonometric representations of various buildings and places based on an initial image that was produced for spatial prototype analysis for own project, The Plug-in School; to better understand the scope of algorithm and how can the creator can fine tune prompts to create meaningful, beautiful and accurate images through a general-purpose GAN.

This project is part of an ongoing design research, conducted at AI Gen Art research group at Columbia Engineering Computer Science department.