Fine-tuning continues training a ready model on a small, domain-specific dataset — your surveillance footage, your medical vocabulary — so it specialises while keeping the broad knowledge it already learned. It is cheaper and faster than training from scratch.