TikTok is finally offering a way to push back against the wave of AI “slop” in your feed, and it’s combining it with some pretty serious behind-the-scenes watermarking technology. Here’s what’s coming next, how it compares to Google’s YouTube approach and Meta’s approach, and why experts say watermarking isn’t yet a magic solution.
Even if you’ve never heard the term “AI slop,” you may have seen it in your video feed or in content shown by friends or news reports. AI-generated videos continue to improve, crossing the “uncanny valley” from looking eerily fake to being quite convincing. Social Video Ground Zero Slop has become a growing part of TikTok, and the company recently announced new tools that allow viewers to lower or raise the flow. TikTok said these features will be rolled out as a test “in the coming weeks.”
This takes advantage of the “Content Settings” slider you can already use to control how much of a particular content type is shown, such as Current Affairs, Dance, Fashion & Beauty, Food & Drink, and more. You’ll soon see a new slider for something called “AI Generated Content” (AIGC). (It’s not yet clear how the controls will be labeled.) Perhaps[プロファイル](lower right corner) > Three parallel lines icon (upper right corner) >[設定とプライバシー]>[コンテンツ設定]>[トピックの管理]I think you can access it by clicking.
read more: Google brings Gemini AI to mobile photo app, offering wild editing options
If you follow the current content control format, move the slider to the left to decrease AIGC and to the right to increase AIGC. Note that the former refers to “few,” not nothing. TikTok does not promise that it will be able to catch and filter out all AI content. But if the new TikTok filters work reasonably well, they could offer the most control ever over AI slop compared to what Google’s YouTube and Meta’s Facebook, Instagram, and Threads currently offer.
How TikTok’s AI filters work
TikTok’s filters rely on an AI-generated content labeling process, which is not foolproof. Video creators using Content Studio are already required to label their AI-generated realistic videos. In addition to its proprietary detection model, TikTok also uses C2PA content credentials, an industry standard that embeds details about how videos and photos were created in the metadata accompanying these files. TikTok says it has already labeled more than 1.3 billion AI-generated videos using these technologies.
This time, an “invisible watermark” will be added. This is a hidden label in AI-generated files that TikTok’s systems can read. The watermark will appear on videos created by TikTok’s own tools (such as AI Editor Pro) and on uploads that already contain C2PA metadata. The goal is to keep TikTok’s AI labels intact even if the video is edited, downloaded, or reuploaded elsewhere.
Google/YouTube and meta efforts
TikTok’s approach is similar to Google’s invisible watermarking technology SynthID, which labels AI-generated images, videos, and audio, and is now being extended to text as well. It is intended to remain intact even after common editing operations such as cropping, filtering, compression, and changing the speed of video or audio. Google is now offering SynthID Detector, which can flag content created with its own tools.
read more: Are you worried about your teen’s TikTok usage? Here’s what you can do
On YouTube, Google is also adding a visible “AI-generated” label to uploads of videos that are AI-generated or heavily edited by AI. Google reserves the right to label any videos it detects that are not labeled as synthetic.
Meta is following a similar path. On Facebook, Instagram, and Threads, photorealistic images created with Meta AI are marked with an “Imagined with AI” badge, as well as invisible watermarks and metadata known as C2PA standards and IPTC. (Meta says there are plans to label the videos.)
For now, the companies are focused on the transparency and disclosure that TikTok also promises, but not on the “robust” filtering capabilities that TikTok promises.
Expert opinion on watermarking
Outside experts and commentators support watermarking, but warn that it is far from perfect. Last week, I discussed C2PA technology in the context of AI-generated images with Jacob Hoffman Andrews, a senior engineer at the Electronic Frontier Foundation. “This is becoming increasingly common among major AI generation tools and will likely help protect against unintentional deception,” he said. “But people who want their AI-edited images to look like they came straight from the camera can simply remove the metadata.” He then named NO C2PA, an online tool he uses to remove C2PA metadata from AI-edited images created in Google Photos.
“Or even easier, take a screenshot,” he added. “Of course, this is well understood within the industry. C2PA is not designed to be difficult to remove. It’s just metadata.”
[Image credit: Sean Captain/Techlicious via ChatGPT]
