Meta boasts that its video AI can “outperform” rivals like OpenAI

• Meta just released Movie Gen, an AI video generator to compete with OpenAI’s Sora.
• Movie Gen can make videos with accompanying audio using a text message. It can also be edited via prompt.
• Meta joined the video generation race later than OpenAI and Google.

Meta released a new AI video generator on Friday, which is also the company’s latest salvo in its battle with OpenAI for AI supremacy.

“Today, we’re excited to release Meta Movie Gen, our groundbreaking generative AI research for media that includes modalities like image, video and audio,” the company said in a press release. “Movie Gen outperforms industry peers in these tasks when evaluated by humans.”

In its press release, Meta called Movie Gen “the most advanced and immersive suite of storytelling models,” including video generation, audio generation, custom video generation, and video editing. The models were trained using publicly available and licensed data, the company said.

With a text message, Movie Gen can create videos up to 16 seconds long at 16 frames per second while reasoning “about object motion, subject-object interactions, and camera movement.” Users can upload a picture of themselves to be embedded in custom videos, and Movie Gen can edit videos with text instructions from the user.

Meta’s sample video shows an underwater perspective of a baby hippopotamus (Moo Deng reference, anyone?) swimming happily in a serene water scene.

Another shows a koala on a surfboard and the accompanying message: “A fluffy koala bear surfs. He has a gray and white coat and a round nose. The surfboard is yellow. The koala bear holds on to the surfboard with its paws. Koala The bear’s facial expression is focused.

With audio generation, users can “create and extend sound effects, background music, or entire soundtracks” up to 45 seconds long, the press release said. An example clip of a snake sliding through a wooded area includes prompt: “Rustling leaves and snapping twigs, with a piece of orchestral music.”

Meta is a little late to the audio and video generation game, as top competitors like OpenAI and Google have already claimed a foothold in the space. OpenAi released Sora, its video generator, in February, and Google followed suit with Veo in May.

The meta, however, gave OpenAI a run for its money in the AI ​​arms race. Although OpenAI’s ChatGPT debuted first and launched the company to global fame, recent iterations of Meta’s Llama model have been well received. Many saw Llama 3.1, which came out in July, as superior to OpenAI GPT-4o, which came out shortly before.

Meta says its new “state-of-the-art models” outperform competitors in human A/B comparisons. For video generation, Meta’s respondents preferred Movie Gen over OpenAI Sora, according to the company’s press release. Meta didn’t share an A/B comparison with Google’s Veo, which also offers sound effects and music, but Meta said in a lengthy accompanying research paper that it believes Google’s video-audio generation models they might be more limited in length than Meta’s. .

Meta, OpenAI and Google did not immediately respond to a request for comment.