FROM SLICES TO SPACES Design ideation on architectural models through AI-generated image sequences

Area of Science:

• Computer Science
• Architecture
• Artificial Intelligence

Background:

• Current methods for stylizing 3D architectural models are limited.
• Exploring diverse design concepts in early stages is crucial for architects.

Purpose of the Study:

• To present a novel AI-driven methodology for style transfer on complex 3D architectural models.
• To enable architects to rapidly explore diverse design concepts and visual outcomes.

Main Methods:

• Converting 3D models into 2D image sequences.
• Integrating sequential slicing, training, and video-guided diffusion.
• Reconstructing transformed 2D sequences back into stylized 3D forms.

Main Results:

• The methodology successfully transforms 3D models based on text, image, or video prompts.
• High-resolution outputs capture exterior and interior spatial relations.
• The AI approach outperforms existing video diffusion platforms in design exploration.

Conclusions:

• The AI-driven style transfer enhances early-stage architectural design ideation.
• This approach facilitates rapid exploration of intricate 3D architectural transformations.
• The method validates the use of 2D image-based AI for complex 3D model manipulation.