Imagen 3

Imagen 3 is our premier text-to-image model, capable of generating images with superior detail, richer lighting, and fewer distracting artifacts compared to earlier versions.

Enhanced Versatility and Prompt Comprehension We developed Imagen 3 to create high-quality images across various formats and styles, ranging from photorealistic landscapes to detailed oil paintings and whimsical claymation scenes.

Imagen 3 comprehends prompts in natural, everyday language, simplifying the process of obtaining desired outputs without complex prompt engineering.

To help Imagen 3 capture nuances such as specific camera angles or compositions in detailed prompts, we enriched the caption of each image in its training data. This improved information allows Imagen 3 to more accurately generate a wide range of subjects and styles. Superior Image Quality Imagen 3 produces visually rich, high-quality images with excellent lighting and composition. It accurately renders small details like the fine wrinkles on a person’s hand and complex textures such as a knitted stuffed toy elephant. Improved Text Rendering We have significantly enhanced its text rendering capabilities, enabling new applications such as stylized birthday cards and presentations.

Built and Deployed with the Latest Safety Innovations Imagen 3 incorporates our latest safety and responsibility innovations, from data and model development to production. We utilized extensive filtering and data labeling to minimize harmful content and conducted evaluations on fairness, bias, and content safety.

We’re deploying Imagen 3 with the latest privacy, safety, and security technologies, including SynthID, a digital watermarking tool that embeds an imperceptible watermark into the image pixels for identification.