Geminis can now recognize videos and sounds created with AI

In recent months, it has become increasingly difficult to identify AI-generated videos, and now Google has added a new feature to Gemini. Users can now upload suspicious videos and ask the AI, “Was this video made by Google AI?” The new feature will accurately scan for invisible watermarks, not only to determine if the video is genuine, but also whether the audio has been tampered with.

According to The Verge, Gemini’s new feature is based on Google’s invisible watermarking technology called SynthID. When you upload a video to Gemini (up to 100MB and 90 seconds), the AI ​​scans each frame and audio line for this digital signature. Gemini’s answer isn’t a simple yes or no; it tells you exactly where, for example, the audio was AI-generated between seconds 10 and 20. Of course, the feature is currently limited to content created with Google’s own tools (like Veo or Imagen). AI Image Recognition with Gemini But can Google’s invisible watermark be removed? Google claims that the SynthID watermark is “subtle and robust,” but experience has shown that watermarks can be removed with heavy editing or compression. The system also can’t currently recognize videos made by competitors like Sora or Kling. Still, embedding the C2PA standard metadata in files is a step forward for transparency in the digital space.

In addition to video recognition, Google has introduced a new way to interact with the Nano Banana imager model. You can now draw directly on a photo you've uploaded or annotate a specific section. For example, instead of typing "color the hat red," just draw a line around the hat and write "red." This feature makes interacting with the AI ​​much faster and more intuitive.