Exploring Generative AI Responsibly with Google
Our recent introductions demonstrate our dedication to developing technologies in a responsible manner, aligning with Google’s AI Principles. With the improved capability to create lifelike images with Imagen 2, we have made substantial investments in ensuring the safety of training data and implemented technical safeguards to restrict the generation of problematic content such as violent, offensive, or sexually explicit material. Additionally, we have incorporated filters to reduce the likelihood of generating images featuring identifiable individuals. Rigorous adversarial testing is also conducted to detect and address potential harmful or problematic content.
All images generated with ImageFX and music produced using MusicFX are embedded with a digital watermark known as SynthID, developed by Google DeepMind. This watermark is imperceptible to the human senses, enabling the detection and identification of AI-generated content. Furthermore, all images created with ImageFX include IPTC metadata, providing additional information to individuals who come across our AI-generated images. We are committed to furthering the development of these and other techniques to enhance the safety and privacy measures of our models.
These tools are currently accessible in the U.S., New Zealand, Kenya, and Australia. Explore these experiments and more at labs.google.
