Pre-rendered CGI has the benefit of time. You can spend more processing power rendering light and effects via ray tracing or more computationally expensive rendering methods. It can take several seconds to render each frame because you only have to do it once. Real time rendering requires trade-offs, and only recently did real time ray tracing for light, shadows, and reflections become possible. Compare a game using traditional rendering (using PBR techniques and a light function called the BRDF) to a game using full ray tracing, and you'll find the ray traced game to be more realistic. With PBR and the BRDF, all code is a kind of approximation for what the scene would look like based on our understanding of science. With Ray tracing, it's closer to a physical simulation of light.<p>Additionally, movies are usually recorded in 24-30fps, while games are rendered at 60-120-240 FPS. People who are not used to 60FPS can find the smoothness jarring (I certainly don't like watching 60fps scenes in movies, but regularly play games rendered at 60hz)
Rendered images can only contain what rendering has been programmed into it. An image of a real scene has millions of bits of detail that are missing from rendered scenes.<p>Example: blades of grass. A rendered scene with blades of grass would need to render thousands or even millions of random blades of grass. Who can justify programming each and every one of those. A real scene contains all those blades of grass in the physics of focusing light.
Real time rendering limits the resources (CPU/GPU available) compared to re-rendered stuff (movies, TV shows, pre-rendered cut scenes). This puts a limit on the complexity of the models or environments in the scene, which makes it look less real.