You can also drop an image onto the panel.
A vision model describes every region of an image with a long list of numbers, its features. Two regions get a high similarity score when their features are large and point in the same direction (their dot product). The lines connect the pairs of regions with the highest scores, so you can see which parts of two images the model considers alike. Thicker, brighter lines mean more similar. Everything runs in your browser, and your images never leave your device.
See also the LaFAM demo, which shows where a vision model looks.