Abstract: ‍ ‍Video is a ubiquitous medium that has found widespread use in various fields. The advent of short-video software such as TikTok has fueled iterative updates in video-related technologies. Video inpainting is a current hot topic in video-processing research. It focuses on repairing the damaged areas of video frames using pixel information within frames and temporal reference information between frames. This technique has broad application prospects， including inpainting videos with corruption， removing objects， and detecting video forgeries. Video inpainting can be traced back to old movie restoration techniques utilized at the end of the 20th century. Typically， the movies were repaired frame-by-frame by professional technical teams. In recent years， some artificial intelligence techniques have been used for video restoration， making it easier to revitalize old movies. Video inpainting techniques can be divided into two categories： traditional and deep-learning-based methods. Their lack of understanding of high-level semantic information makes traditional methods less effective for complex scenes and large missing areas. In contrast， their optimized algorithm framework and improved graphics processor performance allow deep-learning-based methods to achieve excellent results in video inpainting by significantly improving the semantic structure accuracy and time consistency. This paper first briefly reviews traditional video inpainting methods. Then， four types of deep-learning-based video inpainting methods are discussed in detail by analyzing their network structures， parameter models， performances， advantages， and limitations. In addition， the commonly used datasets and evaluation metrics for video inpainting are introduced. Finally， the challenges and prospects of the video-inpainting technique are discussed.