Google LLC today said that it’s making its largest and most capable generative artificial intelligence model to date, Gemini, available for developers and enterprises to build their own apps that take advantage of its sophisticated reasoning capabilities and native ability to understand text, code, audio, images and video.

Gemini comes in three different sizes: Ultra, Pro and Nano. The smallest model, Nano, is available in Android starting with the Pixel 8 Pro phone, and a specialized Gemini Pro model is already running behind the scenes in Google’s AI chatbot Bard. The Ultra model is the largest and most capable version in the set designed for the most complex tasks, but will not be available more broadly from Google until sometime in early 2024.

Starting today, developers will be able to access Gemini Pro through Google AI Studio to quickly and easily and start using it immediately. AI Studio is a free, web-based tool that allows developers to develop prompts that allow for testing and using the model and then provides an application programming interface key that can be exported to an app or into another development environment.

With AI Studio, developers can sign up with their Google account and will receive a free quota, which allows up to 60 requests per minute. Once ready, developers can click on “Get code,” and transfer their work to their development environment of choice such as Vertex AI, Google Cloud’s fully managed AI platform that contains numerous models and customization capabilities.

Within AI Studio and Vertex AI developers get access to Gemini Pro, which is a text-only model with a 32,000-token context window, and Gemini Pro Vision, which accepts text and imagery or video as input and outputs text. The model can operate in 38 languages and is available in 180 countries.

As an example, during a press briefing Google showed off how Gemini Pro Vision could be prompted with images of the inside a home for sale with a query to generate a real estate listing. The model used the prompt and the images to produce a written listing that could appear on any given website. Once the developer had used the imagery and prompt, the underlying code generated by the model could then be exported to an app via an API or Vertex AI for the developer to continue to configure or customize their work.

“We want to make sure that developers can easily get into developing with Gemini at whatever point they’re at,” Jeanine Banks, head of developer relations at Google, told SiliconANGLE in an interview. “That’s why we’re providing multiple different software development kits.”

Software development kits are available for Gemini Pro for developers to immediately start building apps that run on any platform including using Python and Node.js, Kotlin for Android, Swift for iOS and JavaScript. Banks also added that developers can get started quickly with templates in AI Studio that will jumpstart them for common projects.

One of the biggest benefits of Vertex AI, with access to the Gemini models, is that developers and enterprise users can tune and customize the models with company data and augment them to generate responses that fit brand and tone. Instead of a real estate listing that looks like every other post on the internet, the developer can fine-tune the model according to their own personalized writing style or company marketing strategy.

Using Vertex AI models can also be customized using company data, including prompt engineering and reinforced with human feedback. Companies can also augment models with real-time data from public and private databases in order to increase accuracy.

Right now, developers will get free access to Gemini Pro and Gemini Pro Vision through Google AI Studio. Vertex AI developers get access to the same models with similar rate limits, 60 requests per minute, at no cost until general availability early next year. After that, there will be a charge of $0.00025 per 1,000 characters or $0.0025 per image for input and $0.00005 per 1,000 characters for output.

Early next year, Google is launching Gemini Ultra, the largest and most powerful Gemini model for complex tasks. This model is currently available for a select number of users on Vertex AI right now.

In addition to Gemini Pro coming to Vertex AI, Google introduced a massive upgrade to the company’s image-generating text-to-image model with Imagen 2, which is now generally available. Imagen 2 is an upgraded version of Imagen, which is capable of producing photorealistic quality images from text prompts, correctly rendering text inside of images on objects and generating logos on objects.

The ability to correctly produce text within images is a particular challenge for text-to-image AI generators. Often image image-generating AI models have great difficulty producing readable or even correctly displayed text when prompted to do so, but Imagen 2 excels at creating text within images, even written on objects. It can also generate a wide variety of logos drawn atop objects within its generated scenes, including lettermarks, watermarks and emblems for branding purposes.

Image: Google