Safely exploring generative AI with Google
At the moment’s launches prolong our dedication to growing applied sciences responsibly, in alignment with Google’s AI Principles. With the elevated potential to generate photorealistic photographs with Imagen 2, we’ve made vital investments in coaching knowledge security and added technical guardrails to restrict problematic outputs like violent, offensive, or sexually specific content material in addition to making use of filters to scale back the chance of producing photographs of named people. We additionally conduct in depth adversarial testing to determine and mitigate potential dangerous and problematic content material.
All photographs generated with ImageFX and all songs produced utilizing MusicFX are marked by SynthID, a instrument developed by Google DeepMind that provides a digital watermark instantly into the content material we generate. SynthID watermarks are imperceptible to the human eye and ear, however detectable for identification. As well as, all photographs generated utilizing ImageFX embody IPTC metadata, giving individuals extra info every time they encounter our AI-generated photographs. We’ll proceed investing in these and different strategies to enhance the security and privateness protections of our fashions.
These instruments can be found within the U.S., New Zealand, Kenya, and Australia. Go to labs.google to strive these experiments and extra.