SynthID can't identify watermarked images with 100% confidence but can distinguish between images that might contain a watermark and those highly likely to contain one. Despite the growing use of watermarking techniques for generative art, a common watermarking standard remains elusive. DeepMind is considering making SynthID available to third parties in the near future, but it's uncertain if third parties, particularly those developing open source AI image generators, will adopt the technology.
Key takeaways:
- Google DeepMind, in partnership with Google Cloud, is launching a tool called SynthID for watermarking and identifying AI-generated images. The tool is currently only compatible with images created by Google’s own image-generating model, Imagen.
- SynthID embeds a digital watermark directly into the pixels of an image, making it detectable by an algorithm but not by the human eye. The watermark remains in place even after image modifications such as adding filters, changing colors, or compressing images.
- While SynthID cannot identify watermarked images with 100% confidence, it can distinguish between images that might contain a watermark and those that are highly likely to contain one. However, it is not foolproof against extreme image manipulations.
- DeepMind is considering making SynthID available to third parties in the near future. However, it is unclear whether third parties, especially those developing open source AI image generators, will adopt the technology.