Abstract: The techniques disclosed herein enable systems to perform automatic camera rotation for multi-posture devices (e.g., tablet devices, foldable devices, external displays with built-in webcams) irrespective of a current posture of the computing device. This is accomplished by reformatting an image captured by an image sensor using an image signal processer prior to encoding for output. To reformat the image, the computing device determines an angle of rotation for the computing device relative to a default posture. The image is then rotated based on the angle of rotation such that the image appears right side up for the current posture. Subsequently, the image is cropped from its original aspect ratio to a predetermined aspect ratio. Finally, the image is encoded to generate an output stream that can be received by an application. In this way, the system maintains visual consistency even as a user rotates the device.