With the advent of products such as ChatGPT and Dall-E, there has been a lot of buzz of late regarding Generative AI and its impact. I would like to explore the potential impact of Generative AI technology on video-based content creation. I would touch upon what Generative AI for video implies, for how well a machine understands the human visual world in general.

Let’s start with the Generative AI-based video content creation. There are products out there from many startups that help in humanizing text delivery for presentations and blogs. Just imagine how different it would be if this is a video of me talking instead of you reading this blog! With Generative AI, I don’t have to record my video for this purpose. I can simply use a tool that takes my sample picture and creates a video of me speaking this blog out. As you can imagine, this can be used by authors, influencers, marketers or any content creator, to create more engaging content.

While, initially, the applications will be just a talking face, as the technology matures over time, this will grow to creating gestures, actions and movements of the ‘digital human’. Initial applications of this will include a more realistic looking human presence and activities in meta verse, with applications ranging from virtual reality- or augmented reality-based games, virtual travel, virtual offices, virtual meetings, etc. Movie Special FX are today done through animation software. They will likely move to use machine generated sceneries, backgrounds, human movements, etc. What will be interesting is to see how the machine generated images and videos compare to animated ones created with painstaking human work.

Taking all this to its logical conclusion will be to ask the question, “Will AI be able to create a new movie by itself? How will such a movie be?” While we can make a leap of faith to say that AI will be able to create a movie, it is hard to say how such a movie will be. While AI will make strides towards creating realistic looking content, will it be able to create a movie that will win an Oscar?!

While all this crystal ball gazing is interesting pastime, one needs to ask a basic question: Does Generative AI mean that the machine can “understand” what it sees? For, only if one understands what a scene is can one create the scene. Some would argue that the machine does not understand; it simply “extrapolates” from what it has seen before and creates a similar looking scene. But I am not sure. I have seen modern, abstract, art created by a program and it looks like a good modern art created by a human. Where does human creativity come from – we truly create something original or do we extrapolate from what we already know? I don’t know the answer.

If it is indeed true that machines can “understand” a scene, then this understanding can help solve a lot of real-world automation problems, be it in automotive, robotics or security industries. We have read about Tesla Autopilot crash into vehicles in very common scenarios where a human would never crash. With the new technologies that underpin Generative AI, will it be possible to solve such basic “understanding” problems of self-driving cars? Another example is an ambitious requirement that I have seen in Indian Safe City tenders – that a camera identify ‘eve teasing’ using Video Analytics. Knowing what is eve teasing and what is just a bunch of friends talking is beyond current Video Analytics technology capabilities. However, with the use of technologies underpinning Generative AI, will it be possible to make significant progress on such hard problems?

If such problems could be solved by a machine, then we will have “Sentient AI”! Nevertheless, the next few years will be an exciting time for AI in video – fasten your seat belts!