A computer vision model analyzes damage photos in three stages: first, it classifies the vehicle (make, model, body type), then segments the image and identifies visible components. In the third stage, it classifies damage per component by type and severity. The model is trained on millions of real damage images and achieves prediction accuracy comparable to experienced assessors.