Google Launched Imagen 2 AI Image Generator with Text and Logo Creating Capabilities

Google recently launched Imagen 2, the second iteration of its AI text-to-image generator and AI image editor that is said to be the firm’s most advanced yet and can synthesize images, text, and logos. 

This upgraded version of Imagen was previewed in an event in May but is now released to users of Vertex AI on Google Cloud. 

However, Google is still not disclosing the training data used for this model, nor are they allowing an opt-out option for artists to retire their work from potentially being used to train Imagen 2. 

Google Imagen 2 – Improved AI Image Generation

Google had previewed Imagen 2, the second version of their proprietary AI image generator, at their I/O conference back in May 2023, but yesterday, the tech giant released the model on Google Cloud, making it accessible for approved users on Vertext AI, its platform for development and deployment of AI tools. 

Obviously, this is a release aimed more at enterprises. Vertext AI customers are now able to personalize and deploy Imagen 2 to simplify the creation of brand-aligned images, using user-friendly tools, fully managed infrastructure, and integrated privacy and safety features.

Developed with technology from Google DeepMind, Imagen 2 boasts significant improvements compared to its predecessor when it comes to generating high-quality, photorealistic images from natural language prompts. It also introduces new features:

• Improved text rendering in multiple languages – Accurately synthesizing text in visuals is something that many of the first and second-generation AI image generators struggled with, but Imagen 2 seems to handle well

• Creative logo generation – The model can generate brand logos in a variety of styles and even overlay them on existing images

• Descriptive captions – The technology has an advanced understanding of visual content, and you can use it to create detailed captions to describe an image or ask it questions about an image’s content to get accurate answers
• Multi-language prompts – Imagen 2, in its current preview, supports five additional languages to English: Chinese, Hindi, Japanese, Korean, Portuguese, and Spanish. Many more languages will be added through the next year

Security Measures for Responsible AI Development and Use

The company informs that its latest image-generative model is equipped with safety measures to align with Google's Responsible AI principles. 

This includes safety filters to prevent the generation of potentially harmful content and the integration of SynthID, an experimental technology from DeepMind that applies invisible watermarks to its images, claiming resilience to edits like compression and color adjustments, that helps easily identify an image as AI-generated and produced by Imagen. Google provides a tool for detecting these watermarks, but it is not accessible to third parties.

Plus, Vertex AI's indemnification commitment now covers Imagen on Vertex AI, encompassing Imagen 2 and future upgrades of the model powering the service. This unique copyright indemnification approach, which backs up the model’s training data and the generated outputs, is meant to provide customers with assurance when utilizing these generative AI products.

Potential Legal Issues with Google Imagen 2

However, and as TechCrunch appropriately noted in their coverage, Google hasn’t disclosed anything about the datasets used to train Imagen 2. This is especially relevant now that we know that the EU’s AI Act bill –which could become effective in 2025– includes a requirement for general application AI models –like this one– to inform of the origins and contents of its training datasets. 

Equally problematic, Google lacks an opt-out mechanism for artists to remove or prevent their artwork from being used to train Google AI models. And doesn’t have any program to compensate these artists for the use of their work in AI model training, willingly or not—something several other developers do provide. 

That said, Google Imagen 2 seems to have satisfying results in text-to-image generation. If you want to give it a try, you must contact a Google Cloud representative to request to join the Trusted Tester Program.

THE AUTHOR

Ivanna Attie

All About Ivanna

I am an experienced author with expertise in digital communication, stock media, design, and creative tools. I have closely followed and reported on AI developments in this field since its early days. I have gained valuable industry insight through my work with leading digital media professionals since 2014.

• NETWORKING