Artificial intelligent software, also known as AI, is a computer code developed to perform a specific task. AI was first used in war strategy games like Battle Lines, but later it started to be used in software for different purposes. These days, you can find hundreds of games with artificial intelligence, which make use of complex AI to give the player an experience similar to that of playing chess.
AI is one of the most dynamically developing technologies, and it is becoming a more and more functional tool in different industries. Started with analysis and prognoses, AI has passed to creative areas by now, such as composing music, restoring old photos, and even creating videos.
How does it work? AI studies and analyses thousands of ready-made films, audio tracks, photos, etc., and then it applies what it has learned. This way, AI can create something new or edit something that already exists – photos, audio, and videos. As automatic image enhancement has already become an ordinary tool in different photo editing programs, editing videos with AI is a relatively new feature.
One of the first experiments of video editing with the help of AI was in 2016. The Watson supercomputer created a trailer for a film. The computer analysed other trailers and chose scenes for a new trailer from the film based on this information. After that, Watson cut the footage and put fragments it had selected into one video.
Although the result was not perfect, that was an excellent demonstration of the abilities of AI. It might have taken many hours to do the same job for a human, while AI can do it much quicker, and a human only has to make some edits to improve the result and make it smooth and watchable.
From that day, AI, or Artificial Neural Networks (ANNs), became a more common tool in video making, and this tool keeps developing. Now people can use AI to improve the quality of old videos, make them suitable for big screens, and colourise videos. And the results are getting more and more impressive.
For example, modern artificial neural networks can upscale a film from an SD resolution that is only 720×576 pixels to 4K. Luckily, it’s now not the privilege of the ones in the know. Movavi has recently added an AI-upscaling feature to the video converter to help users improve video resolution without quality loss. AI-upscaling can increase the image or video quality up to 4 times and has a huge potential for the future. However, it still needs some improvements in the next few updates, to keep up with the development of AI technologies.
As tech companies are starting to play around with AI features and make them available even for not-so-tech savvy users, these tools are worth paying attention to. Different streaming platforms also start using AI and Deep Learning to stabilize a smooth video delivery and improve the users’ quality of experience.
Deep Learning or a Deep Neural Network-based quality enhancement is a technology of intelligent transmission of a video by using the differing content redundancies. This neural network is still developing; the technology is going to use content awareness and device awareness to personalize the video delivery and improve the users’ experience. Nowadays, people are becoming less and less tolerant to long load time, buffering, and pixelation of videos, which means the technology will be developing, and DNN, together with ANNs, have a chance to become mainstream.
One more proof of the growing popularity of AI-processed remastering is the quality of users’ experiments with it—for example, several successful attempts at upscaling old videos of a YouTuber and developer named Denis Shiryaev.
Shiryaev’s first experiment was the enhancement of a classic 50-second film named “L’Arrivée d’un train en gare de La Ciotat” that was first demonstrated to the public in 1896. Shiryaev used two different programs. Gigapixel has helped to upscale the resolution, and DAIN created new frames and inserted them into the video. Thus, Shiryaev was able to increase the FPS to 60.
As the first try turned successful, Shiryaev made a step further. For the second experience, the developer has chosen a well-known film of New York City in 1911. This time, his aim was not only to upscale the video to the 4K resolution and increase its FPS to 60, but also to colourise the film. Shiryaev used the same instruments to remaster the video as he used before. Still, this time the developer also improved the video’s sharpness with a tool he kept in secret and used DeOldify for automatic colourisation of the film.
Though there were some imperfections in the resulting video, they primarily concern colourisation. Gigapixel and DAIN, together with the unknown tool for restoring a video’s sharpness, showed a fantastic result. The fact that this was not a manual remastering, but an automatic enhancing defies the imagination together with the quality of the resulting video – it is surprisingly watchable and smooth.
According to the results of these experiments, we can say that restoring and remastering old films set several tasks for us at once:
- to enhance the resolution of the original video
- to increase its FPS by generating and adding new frames
- to remove noises from the footage
- to increase the contrast of a video
- to colourise the film if it is recorded in black and white, or if the quality of colours is too low
Neural networks are supposed to learn to solve these problems. For sure, they need some time to train, but they are already showing rather up-and-coming results. That is why AI remains a trending topic for discussions and a promising area for IT specialists