Google DeepMind’s SynthID Watermarks AI Images, Videos, Audio, and Text

Google DeepMind’s SynthID now embeds an invisible watermark in AI‑generated images, video, audio and text at the moment of creation, making it detectable by a special detector but invisible to humans. Launched in August 2023 for Google’s Imagen model, the system has already been applied to more than 10 billion pieces of content as of May 2025 and works across Google’s Veo, Lyria and Gemini models. In May 2026 Google announced a partnership with OpenAI so that images made with ChatGPT, DALL·E, Codex and the OpenAI API also carry SynthID marks, moving the technology toward an industry‑wide standard. The watermark survives common edits like JPEG compression, resizing, colour shifts and light cropping, though strong filters or deliberate tampering can weaken it. A public detector lets journalists and users upload files to see if they carry the mark. Unlike older detectors that look for visual artefacts, SynthID stamps provenance at creation, helping build trust in generative AI as the technology improves. As watermarking spreads, efforts to strip it will grow, so the tool will need continual upgrades. With AI‑generated media spreading fast, will you start checking images for hidden watermarks before sharing or trusting what you see online?

SOURCE: https://techcabal.com/2026/06/25/what-is-synthid-and-how-does-it-detect-ai-generated-images/