All of them need a vacuum because you can't shoot elecrons through air, as mentioned at Video "50,000,000x Magnification by AlphaPhoenix (2022)".
TEM: sample has to be very thin, you get a 2D image. Higher resolution possible.
SEM: sample does not need to be ultra thin, you get a 3D image. Lower resolution possible.
It sees and moves individual atoms!!!
This technique has managed to determine protein 3D structures for proteins that people were not able to crystallize for X-ray crystallography.
It is said however that cryoEM is even fiddlier than X-ray crystallography, so it is mostly attempted if crystallization attempts fail.
By looking at Figure 1. "A cryoEM image", you can easily understand the basics of cryoEM.
We just put a gazillion copies of our molecule of interest in a solution, and then image all of them in the frozen water.
Each one of them appears in the image in a random rotated view, so given enough of those point of view images, we can deduce the entire 3D structure of the molecule.
Ciro Santilli once watched a talk by Richard Henderson about cryoEM circa 2020, where he mentioned that he witnessed some students in the 1980's going to Germany, and coming into contact with early cryoEM. And when they came back, they just told their principal investigator: "I'm going to drop my PhD theme and focus exclusively on cryoEM". That's how hot the cryo thing was! So cool.