Figure 1: We introduce Lifting AutoEncoders, a deep generative model of 3D shape variability that is learned from an unstructured photo collection without supervision. Having access to 3D allows us to disentangle the effects of viewpoint, non-rigid shape (due to identity/expression), illumination and albedo and perform entirely controllable image synthesis. AbstractIn this work we introduce Lifting Autoencoders, a generative 3D surface-based model of object categories. We bring together ideas from non-rigid structure from motion, image formation, and morphable models to learn a controllable, geometric model of 3D categories in an entirely unsupervised manner from an unstructured set of images. We exploit the 3D geometric nature of our model and use normal information to disentangle appearance into illumination, shading and albedo. We further use weak supervision to disentangle the non-rigid shape variability of human faces into identity and expression. We combine the 3D representation with a differentiable renderer to generate RGB images and append an adversarially trained refinement network to obtain sharp, photorealistic image reconstruction * Indicating equal contributions.results. The learned generative model can be controlled in terms of interpretable geometry and appearance factors, allowing us to perform photorealistic image manipulation of identity, expression, 3D pose, and illumination properties.