The future of filmmaking: AI

In recent weeks, I have been writing about AI from the perspective of image production, but my daily work is in film production and animation. The landscape of video production and storytelling is in a state of flux, and with the latest AI technologies, such as the SORA model developed by OpenAI, we are entering a new era. New programs for creating moving images are emerging almost weekly. However, many of these are not yet suitable for high-quality work. And even the best have difficulty producing good quality video that can be used for more than a second per shot. Almost all videos made with current tools also have a slow-motion feel to them.

Despite these shortcomings, I am excited about the development. I believe that all of the problems I have listed above will be solved within this year. In the future, new tools will bring efficiency to productions, and creators will have more time to refine scenes in which real actors perform, for example. Shots that support the storytelling can be created with AI later or even before the actual filming. As creators, we will have more creative opportunities to tell stories. AI programs allow video makers to explore new ideas and create scenes that would otherwise be too difficult or expensive to realize using traditional methods. However, the normal production method will not disappear, and we will continue to shoot on location as well as in virtual sets. AI simply brings many more opportunities to tell better stories.

Inspired by this, I decided to test what the current tools are capable of during my vacation (for the love of the game :). I quickly created a story for Ferrari (a high level of difficulty because creating movement for cars has been very difficult), also using Chat GPT to my advantage, but also my own experience, and iterating the machine-generated script many times. I then voiced the voiceover myself to add more liveliness to some parts. However, I changed my own voice to my idea of the speaker using Eleven Labs. Then I started creating scenes with Midjourney, and 50 believable images were created. The number of images was so large because I knew the challenges of creating videos with AI. And then I created videos from the images using Runway and Pixverse. Many, many, many iterations later, I was relatively happy with the videos (all still have many problems). Then to the editing table, and I decided to try the free CapCut, which was surprisingly good. After editing a one-minute set, I made the appropriate effects and created the soundscape using traditional methods. Here is the result. It is not perfect, and some scenes are more realistic than others. There is some squashing in some elements, but structurally it is starting to come together quite well. After this test, I was incredibly excited about the possibilities of the future.