Artificial intelligence systems that convert text to images are now booming in both capabilities and popularity. And now such a system has appeared in the social application TikTok.
The video platform has recently added a new effect called AI greenscreen. It allows users to enter a text description, which the program will generate as an image. This image can then be used as a video background. This is potentially a very useful tool for content creators.
The output of the TikTok system is quite simple compared to modern text-to-image models such as Google Imagen, OpenAI DALL-E 2 or Midjourney. She creates only rather abstract and swirling images. Other models can create both photorealistic images and complex and cohesive illustrations that look like they were drawn by people.
However, the limitations of the TikTok model may well be intentional. First, more advanced models require more processing power, which would be costly and resource intensive for a company to implement. Second, TikTok has over a billion users, and giving all those people the ability to create photorealistic images of anything they can imagine would almost certainly lead to disturbing results. So instead of naturalistic scenes of nudity or violence, the TikTok system produces only naturalistic colors and vaguely resembling shapes, but not shocking content.
Source: The Verge