Even your TikTok dances aren't safe from AI training

When Alibaba revealed Animate Anyone, its AI tool capable of turning static images of people into moving ones, many speculated that it could put TikTok influencers out of a job. And now, it's been revealed that the tool was indeed trained on date from scraped TikTok videos.

Well-known TikTok creators, including Charli D’Amelio, Addison Rae, Ashley Nocera, and Stina Kayy are cited in a paper detailing the model, which is capable of animating still images and has been shown to produce extremely realistic TikTok-style dance videos based on a single still.

Datasets used to train AI have come under increased scrutiny in recent weeks after a document naming thousands of artists scraped for Midjourney was leaked this month. In the paper describing Animate Anyone, Alibaba claims to have used  “the TikTok dataset” And as 404 Media points out, that dataset, produced in 2021 at the University of Minnesota, inc, includes over 300 scraped videos. Screenshots in the Alibaba paper show reference images of videos from creators like Jasmine Chiswell, Mackenzie Ziegler, and Anna Šulcová.

While Alibaba argues that TikTok content falls under fair use, it's another example of content being scraped for AI without the creator's consent. With the likes of even Adobe drawing the ire of artists for supposedly facilitating copyright infringement, any example of egregious content scraping is coming under scrutiny right now. If even the TikTok dancers aren't safe, is anyone?