Haiper has joined forces with researchers from the University of British Columbia to unveil a new method that transforms photos into spatially consistent 3D objects. This development allows us to view an object from multiple angles, all derived from one image. The challenge lies in understanding the underlying 3D structure of the object from an image and rendering spatially consistent novel new views of it. Our research is based on the hypothesis that a video diffusion model has already learned such spatial consistency. Therefore, our solution is to reformulate the task into a video generation problem and to instruct a video diffusion model with simulated camera motion to imagine the unseen sides of an object.

For a closer look at how we bring single images to life in 3D, check out our work [ViVid-1-to-3].