Our motion brush project aimed to develop an algorithm capable of transforming a single image into a captivating and seamless animated loop, known as a cinemagraph. This cutting-edge algorithm can detect periodic structures within an image, for example the windows of a building or the steps of a staircase. Next, it generates a non-trivial displacement vector field that precisely maps each segment of the structure onto a neighboring segment, following a user- or auto-selected main direction of motion. Hence the name Motion Brush, as the user can create motion in a scene much like as with any other brush in an image editing program.

The resulting displacement field, complemented by appropriate temporal and spatial smoothing techniques, is utilized to warp the image dynamically. This process leads to the creation of frames that seamlessly transition into each other, forming a continuous and mesmerizing cinemagraph. What sets our approach apart is its efficiency, as our cinemagraphs can be generated in under a second on a mobile device.

Through our innovative algorithms, we empower users to breathe life into static images, transforming them into captivating visual experiences. Our cinemagraphs encapsulate the beauty of motion within a single frame, providing an immersive and enchanting glimpse into the world captured in the source image. We developed the algorithm with repetitive patterns in mind, but found out it works very well on other kinds of textures, such as fluid elements: sand, water, sky, snow, etc.

The work on this project was submitted and accepted to the ACM SIGGRAPH conference under the title "Endless Loops: Detecting and Animating Periodic Patterns in Still Images", and was published in the ACM Transactions on Graphics. The submission highlights the novel contributions we have made and the exceptional outcomes of our cinemagraph creation method.