RBM – Roof Box Machine

Fine-tuned Stable Diffusion model for roof box design purposes.

“Communication and storytelling through images and visuals have been a part of human history since the dawn of time. Starting with cave paintings and progressing through pens and paint to pixels on a screen, the techniques and procedures by which visual material is created are continuously evolving. We are currently in the midst of a transitional period into a new era of artificial intelligence in image creation. The technological advancement of clip models and diffusion models has paved the way for fast dataset compilation and AI model training. Integrating image-generating AI into the design process can help with a wide range of jobs, such as rendering, storyboarding and ideation. It is a tool or collaborative partner that has the potential to supercharge the designer’s productivity. The limitations of neural networks is that it can only create an infinite amount of combinations of the training result. If the design project falls outside the scope of the neural network, AI can not assist. This implies that in order to realize the full potential of AI, you need to train on the tasks that you wish to perform. The result of my project is a fine-tuned Stable Diffusion model for the purpose of roof box design. Low-rank Adaptation training has significantly improved the model’s understanding of roof boxes and can aid in the process of designing the visual language of new roof boxes. “

Project info
Stefan Sotra