Google's New AI Tools

On May, 14 Google has announced major updates of their AI technologies during Google I\O Dev Conference. In this article, we will discuss all the models and how to access them. If you don't have access yet, we will explain how to join the waiting list and share the release dates for each AI model from Google.

You can read detailed information about all the updates in our recent article.

Briefly about new Google AI products:

• Gemini 1.5 Pro: An updated version of Gemini Pro with a context length of 1 million tokens, designed for handling extensive text and upcoming support for video processing.

• Gemini 1.5 Flash: A lightweight, fast version of Gemini Pro, optimized for low latency and multimodal inputs, performing better than previous versions in benchmarks.

• Gemini Nano: A super lightweight, multimodal LLM designed for Android and is to be integrated into Chrome, providing efficient client-side processing for various tasks.

• Imagen 3: A new image generation model that excels at understanding long prompts and rendering text, available in multiple versions for different tasks.

• PaliGemma: The first open vision-language model in the Gemma series, designed for image captioning, visual Q&A, and other image labeling tasks.

• Gemma 2: The next generation open source model in the Gemma series, tailored for developers and researchers, offering high performance and efficiency on standard GPUs or TPUs.

• Music AI Sandbox: the first music-to-music model. This is a fully-fledged tool with its own interface that refines existing samples, performs style transfers, and creates variations

• Veo: Google's text-to-video model that understands prompts in text, video, and images, generating high-resolution videos. A direct competitors to OpenAI's Sora.

• Google Gems: Personalized versions of Gemini that users can create for specific roles like a gym partner or coding helper, configured easily with prompts. Gems are similar to custom GPTs by OpenAI.

  
  

Accessing New Google Models

Each model offers unique features and different ways to access them. Here, we'll explain how to access these exciting new tools, starting with the Gemini series, Google Gems, and more. For those models currently unavailable to the public, we will provide the release dates announced by Google, as well as links to join waitlists where such an option is available.

Access to Gemini 1.5 Pro

The upgraded Gemini 1.5 Pro can handle a context length of 1 million tokens, which is about 1.5k pages of text – more than the entire book "War and Peace." Soon, it will also support videos up to 1 hour long.

Gemini 1.5 Pro with 1M token context length is now available with the Gemini Advanced subscription and for developers. You can access it through the chat platform at gemini.google.com. Developers will have a private preview version with a context length of 2 million tokens.

Also, you can use this model via ChatLabs AI platform, with access to other top models including GPT-4o, Llama 4, and Claude 3 Opus under one subscription.

Access to Gemini Nano

Gemini Nano is a lightweight, multimodal LLM designed for Android. It will be integrated into the next version of Chrome, running directly on the device. This integration promises excellent autocomplete functionality throughout the browser and provides developers with simplified APIs for translation, text summarization, and audio transcription. This client-side model eliminates the need for expensive cloud-based LLMs, running efficiently via WebGPU and, in the future, NPU.

You can apply for the preview, with the full release scheduled for Chrome 126, which launches on June 5.

Access to Gemini 1.5 Flash

Gemini 1.5 Flash is a lightweight and fast version of Gemini Pro, designed for low latency. It also has a context size of 1 million tokens and performs better than the previous Gemini 1.0 Pro in benchmarks. This model supports multimodal inputs, providing a quick and efficient experience.

Gemini 1.5 Flash is already available as a public preview.

Access to Veo Video Generator

Veo is Google's text-to-video model, a competitor to Google's Sora. It understands prompts in text, video, and images, generating video in 1080p resolution.

To see Veo in action, watch a preview of collaboration with Donald Glover and his creative studio, Gilga, who used Veo model for a film project.

In the upcoming weeks, select creators will gain access to the feature via VideoFX, a new experimental tool at labs.google. You can join the waitlist now. Veo on VideoFX is limited to users age 18+ based in the US.

Access to Imagen 3

Imagen 3 is Google's new image generation model, which better understands long prompts and renders text. There will be several versions of Imagen 3, each optimized for different tasks, from quick sketch generation to high-resolution images.

The model is not publicly available as for now, however you can join the waitlist to gain access to Imagen 3 as a trusted tester and explore its features through ImageFX. If you are from the US, you should be at least 18 years old.

Access to Music AI Sandbox

Music AI Sandbox is the first music2music model! This is a fully-fledged tool with its own interface that refines existing samples, performs style transfers, and creates variations. Producers of all levels will finally have something new to experiment with.

Currently, the tool is only available to a select group of musicians. We just have to wait for news from Google about the public release dates of this model.

Access to Google Gems

Gems are custom Gemini models, similar to GPTs for OpenAI. Gemini Advanced subscribers can soon create Gems, personalized versions of Gemini. You can make a Gem for various roles, like a gym partner, sous-chef, or coding buddy. Setting up a Gem is easy – just describe what you want it to do and how it should respond.

Google will release Gems soon, and they will be available to users in a few months.

Access to Gemini Gemma

Google has expanded its Gemma AI series with new models. PaliGemma, the first open vision-language model in the Gemma lineup, is designed for tasks like image captioning, visual Q&A, and other image labeling activities.

Gemma 2, set to launch in June, is tailored for developers and researchers needing powerful yet manageable AI tools. The Gemma 27B model offers top-tier performance and runs efficiently on standard GPUs or a single TPU host via Vertex AI.

PaliGemma is available today, and Gemma 2 will be released as open source in June, 2024. The visual LLM PaliGemma 3B is also released, with weights available for immediate use. Get access to Gemma API through HuggingFace.

Conclusion

Google's latest AI models, announced at the Google I/O Dev Conference, offer a wide range of features and capabilities. Gemini 1.5 Pro provides extensive text processing and upcoming video support. Gemini 1.5 Flash is optimized for low latency and multimodal inputs. Gemini Nano integrates seamlessly into Android and Chrome for efficient client-side processing. Imagen 3 excels in image generation with multiple versions for different tasks. PaliGemma and Gemma 2 expand the open-source Gemma series, tailored for vision-language tasks and high performance. Veo generates high-resolution videos from various prompts, and the Music AI Sandbox refines and creates music samples. Google Gems allow for personalized AI roles, enhancing user interaction.

After recent releases from Google AI and OpenAI, it's clear these companies are leading the AI technology race. The fast development of new models and tools is amazing, and we can only guess what powerful automation tools we'll see soon. It's very exciting – let's watch and see what happens next.