We present mav3d (Make-A-Video3D), a method for generating three-dimensional dynamic scenes from text descriptions. Our approach uses a 4D dynamic neural radiance field (NeRF), which is optimized for scene appear