Meta (META) debuted a collection of synthetic intelligence fashions known as “Movie Gen” on October 4, able to producing photorealistic movies up to 16 seconds lengthy, accompanied by sound results and background music. The chosen outcomes are spectacular, however these fashions will not be but obtainable for normal testing.
Movie Gen just isn’t the primary multimodal synthetic intelligence mannequin able to generate video and audio from easy textual promptshowever seems to exhibit cutting-edge capabilities. The researchers answerable for growing this utility declare that it has outperformed rival techniques in human testing.
Movie Gen
According to a Meta weblog put up, Movie Gen is presently able to producing movies up to 16 seconds lengthy, with a price of 16 frames per second (FPS). To put this in perspective, Hollywood movies earlier than the digital age had been historically filmed at 24 FPS to obtain what is known as the “cinematic look”.
Although larger body charges are thought of higher in video video games and different graphics functions, Meta’s 16 FPS just isn’t removed from what can be thought of a skilled cinema picture high quality.
Movie Gen fashions can generate fully new movies based mostly on easy textual cues or modify present pictures or movies to substitute or change objects and backgrounds.
Its most superior contribution, nonetheless, often is the suite’s capability to generate up to 45 seconds of audio with sound results and background musicboth. According to Meta, Movie Gen integrates and synchronizes audio with movement in generated movies.
For analysis solely
Meta is maintaining the bottom fashions behind Movie Gen underneath wraps for now. The firm has not given a timeframe for the launch of the product and has stated that it’s going to require extra safety testing earlier than deployment.
According to a analysis article from Meta’s synthetic intelligence staff:
“Movie Gen’s base set of models was developed for research purposes and needs multiple improvements before being deployed. When we release these models, we will incorporate security models that can reject prompts or generations that violate our policies to prevent misuse.”
