was released.
new updates
greater diversity and prompt understanding
Google designed Imagen 3 to generate high-quality images across a wide range of formats and styles, from realistic landscapes to richly textured oil paintings or fantastical claymation scenes. Imagen 3 can also understand prompts written in natural, everyday language, allowing users to obtain the desired output without needing complex prompt engineering.
To help Imagen 3 capture specific nuances such as camera angles or compositions in long and complex prompts, Google added richer details to the captions of every image in its training data. With better information sources, Imagen 3 can more accurately generate images across various subjects and styles.
Prompt: Shot in the style of DSLR camera with the polarizing filter. A photo of two hot air balloons floating over the unique rock formations in Cappadocia, Turkey. The colors and patterns on these balloons contrast beautifully against the earthy tones of the landscape below. This shot captures the sense of adventure that comes with enjoying such an experience.
higher quality images
The images generated by Imagen 3 are visually rich, high-quality, with good lighting and composition. It can accurately render details such as fine wrinkles on a person's hand and complex textures like the texture of a knitted toy elephant.
Prompt: A large, colorful bouquet of flowers in an old blue glass vase on the table. In front is one beautiful peony flower surrounded by various other blossoms like roses, lilies, daisies, orchids, fruits, berries, green leaves. The background is dark gray. Oil painting in the style of the Dutch Golden Age.
better text rendering
Google has significantly improved its text rendering capabilities, opening up new possibilities for use cases such as personalized birthday cards and presentations.
Prompt: Elephant amigurumi walking in savanna, a professional photograph, blurry background
built and deployed using the latest safety and responsibility innovations
Imagen 3 was built using Google’s latest safety and responsibility innovations, from data and model development to production. Google used extensive filtering and data labeling to minimize harmful content in the dataset and reduce the likelihood of harmful outputs. Google also conducted red team testing and evaluations covering topics such as fairness, bias, and content safety. Do you remember the black-skinned Musk?
— which embeds digital watermarks directly into the image pixels so they can be detected but are imperceptible to the human eye.
Ads, etc.