Keywords
photography, CLIP, aesthetics

Technologies
AI, CLIP

The application integrates a multimodal AI model (CLIP) with a neural probe to provide structured feedback on formal image qualities such as lighting, composition, and visual clarity. It is designed as a tool for systematically guiding users’ attention toward objective visual properties, enabling controlled investigation of how such prioritisation influences aesthetic evaluation.

My individual contribution was the development and deployment of the core machine learning model: a CLIP-based neural probe trained on the AVA dataset of 255,508 crowd-rated photographs, incorporating architectural refinements including a residual projection layer, learnable temperature scaling, and a pairwise ranking loss.