by Meta announced on Thursday the launch of Make-A-Video, an AI video generator that can generate videos based on input text or image prompts. Similar to AI image generators DALL-E and Stable Diffusion, it can even modify existing videos. Not open for use.
In addition, Meta also showed examples of Make-A-Video, including “a young couple walking in the rain” and “a teddy bear painting a portrait”; it also generated a swimming video from a static turtle picture through AI; And concatenate two photos into a video.
Image credit: Meta
Instead of using labeled video data (such as subtitles to describe action content), Meta uses image synthesis data (training still images with subtitles) to train the Make-A-Video model, and applies unlabeled video data training, so that Make-A- Video learns the perception of time and space on text or image prompts, and can display and predict subsequent content in a short time.
However, Meta has not announced when Make-A-Video will be open to the public, but admitted that generating lifelike videos through AI will pose a danger to society, saying that a watermark will be added to all videos generated by Make-A-Video for the public to identify that the video is not Real shot.
source:Ars Technica
—
unwire.hk Mewe Page : https://mewe.com/p/unwirehk
