TikTok lets users dial back AI videos, tests invisible watermarks

Tired of stumbling across photorealistic videos on TikTok only to realize ten seconds later that it was hallucinated by an algorithm? You're not alone, and TikTok is finally giving you a dial to tune it out.

The company announced it is testing a new control that lets users adjust how much AI-generated content (AIGC) appears in their For You feed, alongside new invisible watermarking tech to catch sneaky re-uploads.

Dialing back the synthetic feed

The new control lives inside TikTok's Manage Topics menu, right alongside existing sliders for topics like Dance, Sports, and Food.

You cannot block AI content entirely. Instead, TikTok lets you dial the volume up or down based on your personal tolerance, much like you would for recipe videos.

The timing makes total sense. Following OpenAI's Sora rollout and Meta's launch of its dedicated AI feed Vibes in September, hyper-realistic generated clips have flooded mainstream social channels, especially around historical deep dives and celebrity parodies.

Watermarks you cannot easily strip

Detecting AI content is notoriously hard once a video gets downloaded and re-uploaded, because metadata like C2PA credentials usually gets wiped in the process.

TikTok has already labeled over 1.3 billion videos, but it needs stronger glue. It is now testing invisible watermarks that are embedded directly into the pixels of content made with tools like AI Editor Pro or uploaded with C2PA tags.

Because only TikTok’s internal detection models can read these watermarks, bad actors cannot simply strip the metadata with a quick screen recording or basic trim.

To back this up, TikTok is also setting up a $2 million AI literacy fund to pay groups like Girls Who Code and 20 other experts to flood the feed with educational safety content.

Why it matters

We think TikTok is taking a far more realistic route than its competitors. Instead of cordoning off synthetic media into a ghost-town feed like Meta did with Vibes, it accepts that AI is already everywhere and gives users the steering wheel.

For everyday scrollers, this is an immediate win. If you are sick of uncanny valley history lessons, you will soon be able to dial them down without breaking the rest of your algorithm.

However, there is an obvious catch: proprietary watermarks solve the problem for TikTok, but not for the open web. Because only TikTok can read its invisible tags, the broader internet is still stuck playing cat-and-mouse with synthetic spam.