Top

Movie Gen: why Meta’s video AI is different from others

There can be no social media without AI, especially when it comes to those who invented social media. It was only a matter of time, also because the company had already started investing a lot of money in the development of AI solutions in the past. So, it is no surprise that Meta is about to enter the most important sector for the present and future of tech companies in a big way. It will do so with the launch of Movie Gen, a new model for generating high-quality video and audio from multimodal prompts.

Cutting to the chase, the main feature of the future Meta AI is the creation of video via text description. Movie Gen exploits textual input to generate long, high-definition videos with different aspect ratios. And this is a novelty compared to what we have seen so far. But it is other peculiarities that have attracted interest in what Meta is doing because with Movie Gen you can create customised videos. If you upload a personal photo and add textual prompts, the template is able to generate highly customised (and thus unedited) clips that preserve human identity and movement. Movie Gen can also work on existing videos, transforming them according to textual input to complete an edit in the desired style.

It should also be borne in mind that Meta has also developed Movie Gen Audio, the version of the template dedicated to sound that, on textual input, allows background music, sound effects, or soundtracks to be generated with specific tone, rhythm and style, in line with the video.

How it works

Not yet accessible to the public — because it is restricted to a few users and exceptional partners in the film industry, such as the production company Blumhouse, known for transforming horror film production over the past 20 years — on a technical level, Movie Gen is a transformer model powered by 30 billion parameters and capable of producing, for now, videos of up to 16 seconds at 16 fps and 10-second movies at 24 fps. Movie Gen Audio, on the other hand, was powered by 13 billion parameters and can generate content of up to 45 seconds at 48 kHz, with ambient music and sound effects synchronised with the video produced in relation to the textual request.

Meta claimed to have trained Movie Gen on a combination of licensed and publicly available datasets, including 100 million videos, one billion images and one million hours of audio files. Apart from the numbers, however, Mark Zuckerberg’s company provided no further details.

It is clear that Movie Gen is Meta’s answer to OpenAI’s Sora and Runway, but Meta itself speaks of a superior solution to those of its competitors. This is because the template can make localised changes by adding, removing, or replacing one or more elements, but also global changes by changing the background or style of the content. It also makes it possible to create customised videos almost instantaneously by changing an image and text input, generating clips that are striking for the naturalness of the movements. At least, looking at the examples Meta has released so far.

Meta Movie Gen
Meta Movie Gen

Movie Gen to monetise AI investments

Although it is a point that comes up every time AI and professions are discussed, not everyone was happy to discover Movie Gen’s abilities. Meta is aiming to make its model a reference for Hollywood filmmakers and insiders, and it is not the only one to bet on this sector, of course, causing concern and protests among seventh art workers. There are so many jobs within film productions that risk being skipped, so, as already happened in previous months with the strikes of actors and screenwriters, many technicians are already on the warpath, looking at the future impact AI may have on the transformation of cinema. For its part, Meta said that AI-generated videos will be watermarked to avoid problems with copyright and deepfakes, two threats that are set to grow exponentially when AI text-to-video solutions become affordable.

Changing the perspective, Movie Gen could be the breakthrough Meta has been looking for for years, as it can create, edit, enhance and perfect commercials. A way to monetise the huge investments in the development of AI solutions, as well as, in the long run, a quantum leap to secure revenues that, in Meta’s case, since the birth of Facebook, have been linked to advertisers’ tastes, trends and budgets.

However, Facebook will not be the platform that will inaugurate Movie Gen because as announced by Adam Mosseri, the AI model will be posted first on Instagram, where it has the potential to change the impact, spread and popularity of Reels. However, we will have to wait, as there is currently no release date, apart from the certainty that Movie Gen will be the innovation of 2025.

Alessio Caprodossi is a technology, sports, and lifestyle journalist. He navigates between three areas of expertise, telling stories, experiences, and innovations to understand how the world is shifting. You can follow him on Twitter (@alecap23) and Instagram (Alessio Caprodossi) to report projects and initiatives on startups, sustainability, digital nomads, and web3.