Much like DALL-E 3, visitors to imagine.meta.com can provide a prompt, and Meta’s AI will offer several options. The image up top is a selection Meta AI created. If you’re wondering how Meta trained its AI to generate images, the company says it used 1.1 billion publicly visible Facebook and Instagram images.
That raises some questions about consent: after all, many of these images may have been uploaded before AI in its current form existed, and just because someone wanted to share their artwork with the world doesn’t mean they intended for it to train programs to create similar art. Permission to use art and written works for training AI is an ongoing question with evolving AI and has led to several lawsuits that are still working their way through the courts.
For now, the generated art Meta AI creates isn’t watermarked in a way that can’t be hidden. But Meta says it will roll out invisible watermarks soon that shouldn’t be so easy to cover up.
Leave a reply