A scene presented by a headset is adjusted to correct for distortion from optical errors of an optics block in the headset. To correct for the distortion, the scene is pre-distorted when presented based on previously modeled distortion of the optics block, so distortion from the optics block corrects the pre-distortion. To model the distortion, the headset displays calibration image including features and images of the calibration image are captured from multiple positions. Differences between locations of features in the calibration images and locations of corresponding features in captured images of the calibration image are identified and a distortion correction is determined based on the differences.