Compass Control: Multi Object Orientation Control for Text-to-Image Generation

Personalization with 3D Orientation Control Few unposed Input Images 'A photo of V* car in front of the leaning tower of Pisa in Italy' 0.523 1.047 2.617 3.665 3.141 0.0 1.047 2.094 5.235 3.123 'A photo of a mother walking with a pram on a snowy street, festive Christmas lights, beautiful winter evening scene' 0.675, 0.725 0.80, 0.60 * equal contribution. † work done during an internship at VAL, IISc tion. In this work, we address the problem of multi-object orientation control in text-to-image diffusion models. This enables the generation of diverse multi-object scenes with precise orientation control for each object. The key idea is to condition the diffusion model with a set of orientationaware compass tokens, one for each object, along with text tokens. A light-weight encoder network predicts these com-This CVPR paper is the Open Access version, provided by the Computer Vision Foundation. Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore.