Google has introduced a verification feature for its Gemini AI, allowing users to determine if a video was generated or edited using Google’s AI models. Users can inquire by asking, “Was this generated using Google AI?”
The Gemini AI scans both the visuals and audio of the video for a proprietary watermark known as SynthID. The verification response will include specific timestamps indicating where the watermark appears in the video or audio, rather than simply providing a yes or no answer. This feature follows the rollout of similar capabilities for images back in November, which also focused on content created or modified with Google AI.
Concerns exist regarding the durability of watermarks in AI-generated content. Previous experiences, such as with OpenAI’s Sora app, highlighted that some watermarks can be easily removed. Google describes its watermark as “imperceptible,” though questions remain about how easily it can be eliminated or how effectively other platforms will be able to detect SynthID information and label content as AI-generated. While Google’s Nano Banano AI image model uses C2PA metadata, the broader challenge persists with the inconsistent tagging of AI-generated materials across various social media platforms, allowing potential deepfakes to evade detection.
Gemini has the capacity to verify videos that are up to 100 MB in size and 90 seconds in duration. The feature is accessible in all languages and locations where the Gemini app is available.
Source: https://www.theverge.com/news/847680/google-gemini-verification-ai-generated-videos