3D generative AI has enormous potential for industry, but its quality remains insufficient for professional use. We present Emu3D, a significant advancement in text-to-3D which produces faithful, high-quality meshes with full material control. Compared to prior works that bake shading in the 3D object, we output physically-based rendering (PBR) materials, supporting realistic relighting.Emu3D generates first several views of the object with factored shaded and albedo, and then reconstructs colours, metallicity and roughness in 3D, training the reconstruction network via a deferred shading loss. The latter models shape more reliably by outputting directly a sign-distance function and using a corresponding loss, which, implemented in fused kernels, are highly memory efficient.After mesh extraction, a final texture refinement step operating in UV space significantly improves sharpness and details. Emu3D achieves 17% improvement in Chamfer Distance and 40% in LPIPS over the best concurrent work for few-view reconstruction, and a human preference of 72% over the best industry text-to-3D tool, including those that support PBR, with comparable inference time.