Safely exploring generative AI with Google

Today’s launches extend our commitment to developing technologies responsibly, in alignment with Google’s AI Principles. With the increased ability to generate photorealistic images with Imagen 2, we’ve made significant investments in training data safety and added technical guardrails to limit problematic outputs like violent, offensive, or sexually explicit content as well as applying filters to reduce the risk of generating images of named individuals. We also conduct extensive adversarial testing to identify and mitigate potential harmful and problematic content.

All images generated with ImageFX and all songs produced using MusicFX are marked by SynthID, a tool developed by Google DeepMind that adds a digital watermark directly into the content we generate. SynthID watermarks are imperceptible to the human eye and ear, but detectable for identification. In addition, all images generated using ImageFX include IPTC metadata, giving people more information whenever they encounter our AI-generated images. We’ll continue investing in these and other techniques to improve the safety and privacy protections of our models.

These tools are available in the U.S., New Zealand, Kenya, and Australia. Visit labs.google to try these experiments and more.