Earlier this year, Google introduced Imagen 3, its highest-quality text-to-image model. This newest iteration comes with several improvements to the quality and fidelity of the images it creates. Google claims that Imagen 3 can create images with an incredible level of detail, producing photorealistic, lifelike images with fewer distracting visual artifacts compared to Imagen 2.
Google recently rolled out its Imagen 3 model to all Gemini users worldwide. With this availability, both Gemini and Gemini Advanced users can simply describe the images they want to create, and Gemini will respond with the AI-created image. However, only Gemini Advanced users can create images featuring people.
If you are not satisfied with the image created, you can further refine it by adding more details to the prompt. The more specific you are about the details in the image, the better Gemini can create images based on the prompts. This is because the new Imagen 3 model better understands natural language and the intent behind the prompt, and can even incorporate small details from longer prompts.
Imagen 3 is also very good at rendering text, which has been a challenge for most current-generation image generation models. This allows users to create personalized birthday messages, places with name boards, and more.
Imagen 3 is also available on Vertex AI for developers. Developers can choose between the regular Imagen 3 and Imagen 3 Fast, offering them the flexibility to optimize for quality or latency, depending on their application requirements. The Imagen 3 model can create high-quality images with natural lighting and increased photorealism, while Imagen 3 Fast is suitable for creating brighter images with higher contrast. Google claims that developers can see a 40% decrease in latency with Imagen 3 Fast compared to Imagen 2. Also, developers can configure Imagen 3 to create images in different aspect ratios, including 1:1, 9:16, 16:9, 3:4, and 4:3.
You can try out the Imagen 3 model on Gemini here.