(Credit: Meta)
Meta is showing off a new AI video generator that could rival OpenAI's Sora or Google's Veo.
Meta says Movie Gen can create high-quality (1080p) 16-second videos with sound from a text prompt, though it did not provide a release timeline.
(Credit: Meta)Sora promises 60-second silent clips, while Google touts "over 60-second" clips with Veo but has not specified the audio capabilities. Neither video generator is available yet, though a Veo-powered video background generator is coming to YouTube Shorts later this year.
Movie Gen's sound capabilities come from a separate AI model, dubbed Movie Gen Audio. It can "follow a text prompt" and create "high-quality cinematic sound effects and music synchronized with the video output," according to a Meta research paper.
Movie Gen is targeting professional and amateur filmmakers, and says the model can help them with "multiple tasks: text-to-video synthesis, video personalization, video editing, video-to-audio generation, and text-to-audio generation."
Meta trained the model on "internet scale image, video, and audio data," and the demo video features a pygmy hippo, a reference to the social media craze surrounding Moo Deng, a hippo recently born in Thailand.
Impersonating Moo Deng is a clever way to reveal another core capability. Movie Gen can make videos of anyone (human or hippo) from a static photo. Content creators can theoretically crank out AI-generated videos of themselves with a simple text prompt.
(Credit: Meta)Without proper guardrails, such as restricting video generation of public figures and politicians, this technology could lead to the rapid spread of hyper-realistic misinformation.
Meta did not mention any restrictions. The announcement seems focused on assuring filmmakers that "generative AI isn’t a replacement for the work of artists and animators" and insisting that it's working "closely with filmmakers and creators to integrate their feedback."
OpenAI says it is working on guardrails for Sora ahead of a potential release later this year. In February, the company promised to take "several important safety steps [like] working with red teamers—domain experts in areas like misinformation, hateful content, and bias—who are adversarially testing the model."