Most wedding shoots end with 3,000-8,000 photos. But only 200–300 actually make the final album. Someone has to manually look through every single photo to pick the best ones. That can take hours or even days. So I built a small AI pipeline to solve this. Here’s how it works. First, every image gets converted into a visual embedding using CLIP, a model that understands images and their meaning. Then the system: • removes duplicate or near-duplicate shots • scores image aesthetic quality • groups similar photos into clusters • and finally asks an LLM to rate storytelling value Because a technically good photo is not always the one that captures the emotion. Research shows that combining aesthetic scoring + composition + semantics can approximate how humans judge photos. In the end, the pipeline automatically produces: Top 250 photos ready for album selection. From thousands. In minutes. This was a fun project because it shows something interesting: AI is not just about generating images. It can also help us choose the meaningful ones.

Wedding Album Shortlister

From Days to Minutes, choose photos efficiently

From Days to Minutes, choose photos efficiently