Like Dall‑E and Midjourney, it uses machine learning algorithms (and a vast database of online artwork) to turn written prompts into fantasies.
On Thursday, Meta CEO Mark Zuckerberg unveiled a more animated “make-a-video” of “make-a-scene.”
As the name suggests, Make-a-Video is “a new artificial intelligence system that allows users to turn text messages into short, high-quality video clips,” Zuckerberg wrote in a Meta blog post on Thursday.
ing. Functionally, Video is similar to Scene in that it combines natural language processing and generative neural networks to convert non-visual cues into images, only in a different form of content.
“Our intuition is simple: we learn from text-image pair data how the world looks and how it is described,” Meta researchers said in a research paper released Thursday morning.
Learn how the world works from unsupervised video sequences.” By doing this, while maintaining the “breadth” (aesthetic diversity, fantasy expression, etc.) of the current image generation model, the time required for training the video model is shortened, and the paired data of text and video is unnecessary. I was able to do it.
Like most of Meta’s AI research, Make-a-Video is published as an open source project. “I want to think about how to build new generative AI systems like this,” Zuckerberg said.
“We are openly sharing this generative AI research and its results with the community and soliciting feedback.” keep evolving. ”
As with all generative AI on the market, make-a-video has plenty of opportunities for abuse.
The research team assumed the possibility of malicious behavior and removed NSFW images and harmful phrases from the Make-a-Video training dataset in advance.