DEPTHCRAFTER: CONSISTENT LONG DEPTH SEQUENCES FOR OPEN WORLD VIDEOS

Summary from the TLDR AI newsletter

By using Stable Diffusion video as a prior model, researchers were able to generate extremely precise depth information without using any camera pose or optical flow information.