StereoSpace turns a single photo into a full 3D-style stereo pair without ever estimating a depth map.
Diffusion models sometimes copy training images too closely, which can be a privacy and copyright problem.
TreeGRPO teaches image generators using a smart branching tree so each training run produces many useful learning signals instead of just one.