The vast majority of CGI that you're used to seeing is 3d modeled stuff. That approaches "make a moving image" by simulating an environment, physics, and motion.

"Deepfake"-style CGI is fundamentally different in approach. It does not have a physics engine or anything like that. It is relatively very new to the scene; 3d-model CGI is nearly 50 years old.

This makes it difficult to extrapolate from one to the other.

Human no-name actors are cheap; but physical camera work of any kind is not cheap, and human celebrities are very expensive. A team working on full-CGI scenes isn't just replacing the actors for those scenes - it's replacing the makeup, costume, camera, stunt, set preparation, prop, lighting, safety, animal handling, etc. teams.