Listen to the audio version of the article
Last week, Google officially announced Gemini, its most advanced AI model. Available in three variants – Ultra, Pro and Nano – Gemini adapts to different needs. The main difference between the three versions is the size of the machine learning model. Gemini Ultra is the largest, with 1.6 trillion parameters, followed by Gemini Pro with 100 billion parameters, and Gemini Nano with 10 billion parameters. A larger model size means that Gemini Ultra can learn more information and generate more accurate and creative results, but also that it is more expensive and requires more resources to run. Gemini Pro is a good choice for most applications since it is still large enough to generate high-quality results, but requires much less power than Gemini Ultra. Gemini Nano, on the other hand, is the best choice for applications that require a smaller and less expensive model. It is still capable of generating good quality results, but may not be suitable for applications that require maximum accuracy. While the Nano version is already in use in Android’s Pixel 8 Pro, a special version of Gemini Pro powers Bard and the ultra version is coming soon. New this week is that Gemini Pro is now accessible to developers and businesses.
Main Features of Gemini Pro
Gemini Pro offers a 32K window for text, and future versions feature an even larger window. Since it has just been presented, Google has thought to make it very interesting by offering free use within certain limits and very competitive prices for those who need use, I mean. Furthermore, there are many features that make it potentially easier to use than other solutions such as calling external functions, support for 38 languages and a complete development environment that can be fully interfaced with the most used IDEs.
Gemini Pro Vision e SDK
A technically unexplored aspect of Gemini Pro is its multimodal endpoint, Gemini Pro Vision, which processes both text and images, providing text output. To facilitate application development, SDKs are available for various programming languages, including Python, Android (Kotlin), Node.js, Swift and JavaScript, allowing the creation of versatile and adaptable applications.
Google AI Studio and Vertex AI
Google AI Studio, an all-web development tool, allows developers to quickly create prompts and get an API key for app development. The fact that it offers a free quota of 60 requests per minute makes it ideal for experimenting and creating AI-based applications that are not expected to be used too intensively. For more advanced projects, developers can upgrade to Vertex AI on Google Cloud, which offers additional features for customizing Gemini, with full control of the data and additional security, privacy and compliance benefits. The development environment is focused on obtaining AI-based apps with minimal effort, a goal that is common among companies offering their own models, and the fact that development tools can be integrated with the most widespread ones in the market is certainly a point in favor of Google’s solution.
Imagen2 – new features that make a difference
Also integrated into Vertex is the new version of Imagen which produces high quality images starting from text prompts. In addition to this now widespread function in many AI products, Imagen 2 has the advantage of being able to create images with text specified in the prompt. For example, we can decide the writing on a cup or, as seen in the Google example, the brand of toothpaste on the tube. In fact, it is now possible to ask it to generate a complete company logo and then overlay it on generated images. An interesting and cheap AI development environmentThere is no shortage of AI models to experiment with in this period, but Google’s model offers some simplifications in the development phase that could be useful. Furthermore, the presence of a rather generous free plan and low costs for more intensive use make it very attractive at a time when it is not easy to choose which technology to embrace.