Meta works hard to develop AI models that stand out from their competitors. His latest discovery: a family of basic models called Movie Gen, including a text-to-video model. This model is currently targeting the research world, and could be the next revolution in video creation. Meta says the model was trained on a set of licensed and publicly available datasets.
The Transformer model contains 30 billion parameters and has the ability to create videos up to 16 seconds long at 16 frames per second. In its assessment, the parent company of Facebook and Instagram Ensures that Movie Gen outperforms similar industry models at these tasks.
Audio generation model with 13 billion parameters
Regarding vocal capabilities, another model of voice generation has been developed. Featuring 13 billion settings, it can capture video clips, optional text prompts, and output high-quality audio for up to 45 seconds, including ambient sound, sound effects (foley), and background music, all synchronized with the video content. Additionally, Meta offers audio stretching technology that can create consistent audio for videos of different lengths.
Meta Movie Gen is on the scene! Our innovative research in AI allows media to:
– Convert text to video
– Create a personal video
– Edit video accurately
-Create soundEven though it’s just research today, we can’t wait to see all the ways people are enhancing… pic.twitter.com/I4Bq9if3eK
– Meta (@Meta) October 4, 2024
In all, Movie Gen allows, through text prompts, to convert text to video, create custom videos (from a single image), produce video editing and even create audio. Features that are reminiscent of Sora, the text-to-video tool developed by OpenAI and published at the beginning of the year.
Sora, Arrested in Development
However, since posting several impressive videos, the startup has given no further sign of the sustainability of its tool. “Today, Sora is available for Red Teamers to evaluate critical aspects regarding risk or risks. We also give access to a number of visual artists, designers and filmmakers to get feedback on how we can develop the model so that it is more relevant and useful for creative professionals, and we share our progress.” In our research early so we can start working with people outside of OpenAI.can we read on the page dedicated to the tool.
Will OpenAI take down Sora?
The startup behind ChatGPT has introduced a number of upgrades, including a Her typical family is o1. It has also equipped its AI assistant with a voice function – something that is long overdue. Likewise, we expected an update regarding the possible public release of Sora: the truth is that the company does not provide any news on this topic.
Meanwhile, Tim Brooks, head of Sora research at OpenAI, announced that he would be abandoning ship and moving to another well-known company: Google. “I’ll be joining Google DeepMind to work on creating videos and simulating the world! I’m looking forward to collaborating with such a talented team. I’ve had an amazing two years at OpenAI creating Sora,” Can we read on X. Does his departure mark the end of Sora’s Tool adventures? Nothing is certain.
Selected for you
