Synthaesthetic Portrait Robot
A multimodal generative AI system for robotic painting that converts portraits into physical artworks using vision-language models, diffusion generation, and robotic path execution. Featured installation at SuperAI 2025.
NeurIPS 2025 publication. Featured installation at SuperAI 2025 Singapore, viewed by 7,000+ attendees.
Developed a multimodal pipeline converting portraits into robotic paintings using generative models. The system integrates vision-language models, diffusion generation, and robotic path execution into a seamless creative pipeline.
Technical Pipeline
The system chains together multiple AI models:
- SAM2 for precise portrait segmentation
- Florence-2 for vision-language understanding and scene description
- Stable Diffusion with LoRA for stylized image generation
- ControlNet for structural guidance during generation
- SVG Vectorization for converting generated images into robot-executable paths
Impact
Featured as an installation at SuperAI 2025 in Singapore, where it was viewed by more than 7,000 attendees. The work was published at the NeurIPS 2025 Workshop, contributing to research on human-machine creative collaboration.