I presented our single-image 3D morphology prediction work at IMC21, the 21st International Microscopy Congress, held at ACC Liverpool from 31 August to 4 September. The method tackles the inverse problem of electron microscopy: given one STEM image of a supported nanoparticle, recover the 3D atomic structure that produced it. We use an image-conditioned point cloud diffusion model, and it works for particles in arbitrary orientation, without needing a tilt series. Good discussions with people working on tomography and in situ imaging, and a nice reminder of how broad the microscopy community is. Paper in npj Computational Materials.