The reason the model in your example looks like Keanu Reeves is because it is literally an image of Keanu Reeves mapped to some geometry that echoes the general silhouettes of his head. This is a pretty common thing seen with apps trying to reconstruct a very complex object (the human head) based on very little data (a photo).
If you’ve ever made a character in a video game that offers a fancy character creator (like, say, Black Desert Online) imagine the process to achieve something that looks like Keanu Reeves. In a character creator you are given generic geometry and you use the controls to sort of dial in the look you want. The geometry itself deforms but never changes on the mesh level. The same sort of thing is happening with something like Facebuilder (afaik). Using a generic “human” mesh to block in basic features…but then it relies heavily on mapping a photo over that mesh to achieve a “likeness”.
What you don’t get are all the little details to push something in to the realm of realism. In this case, the materials and textures that make skin and eyes look alive.
If you want a fast, out-of-the-box solution to achieve something more realistic the only thing I can think of is Metahumans. You can’t just upload a photo of Keanu Reeves and have a model of Keanu Reeves pop out the other end but, honestly, you couldn’t legally use a likeness of Keanu Reeves in any significant way anyway. You can’t use Metahuman models in Blender (per Epic’s TOS) but you can use them in Unreal which might be the sort of solution you’re looking for.
