Meta launches an AI that turns any person into a 3D object
This is a subtitle for your new post
Meta has created an AI that can recognize people and then reconstruct them in 3D.
The company announced SAM 3, the latest member of the segmentation model family that is capable of identifying elements in images and video. The novelty this year is a variant called SAM 3D, which not only recognizes people and objects, but also reconstructs them in 3D to use them in videos or export them as separate models.
According to a post on Meta’s website, SAM 3 can recognize objects in images and videos and track them. This is very useful for video editors, since they only need to type the name of the object in the prompt box and the AI will take care of selecting it for modification. Although the company had previously developed earlier versions of this model, they were not able to handle detailed descriptions to segment the content.
To test SAM 3, Meta has launched an application called Segment Anything Playground, where users can upload videos and images to crop and add effects. The app includes some examples with predefined prompts, although you can upload your own content to verify that it works. The editor is somewhat basic and does not compare to what we see in dedicated applications such as Premiere, but it offers a glimpse of what can be achieved.
As we see with Adobe, Meta’s AI breakthroughs will be reflected in its products. The company said they will use SAM 3 to add special effects inside Edits, the CapCut-style tool for creating short videos on Instagram and Facebook. The segmentation model will also be applied in AI-generated videos within Vibes in the not-too-distant future.
Don't Forget To Share
