Google Makes Logo Design & Text Generation Instant With New Imagen 2 AI Tool
By Mikelle Leow, 14 Dec 2023
Images via Google
Your next branding designer might not be a person. Google has unveiled Imagen 2, the latest iteration of its artificial intelligence model that transforms text into images. Available to select Google Cloud customers, the technology is developed with the expertise of Google DeepMind. Imagen 2 not only polishes its predecessor's image quality but also adds a dash of creativity with features like rendering text and logos.
The tool is now more adept at producing higher-resolution, lifelike images that mirror their respective input prompts.
Image via Google
More than that, it’s also a tool for professional creativity, being able to design logos or add catchy text to advertisements—an ability that many rivaling AI art generators struggle with. From emblems to abstract logos, Imagen 2 can overlay these designs on a variety of products, including clothing and business cards.
Image via Google
Image via Google
Imagen 2 is also multilingual and can render text in a multitude of languages like Chinese, Hindi, and Spanish, with plans to add even more in 2024. This opens doors for a global user base to explore their creativity without language constraints.
To smarten the text-to-image model, the tech giant enriched its dataset with detailed image captions, enabling it to grasp a wider array of user prompts and understand the intricate dance between words and images. Further, “we trained a specialized image aesthetics model based on human preferences for qualities like good lighting, framing, exposure, sharpness, and more,” its AI team adds.
Image via Google
With its editing features like inpainting and outpainting, users can integrate new elements into existing images or expand them beyond their original boundaries. Google plans to integrate this technology into its Vertex AI platform soon.
To address the growing concerns over image authenticity, Imagen 2 incorporates SynthID, a DeepMind-developed method, to embed resilient, invisible watermarks in its creations. These watermarks stand firm against edits and require a specific Google tool for detection.
Amid debates over the use of publicly sourced data for commercial AI models, Google remains tight-lipped about the data fueling Imagen 2, TechCrunch points out. With that in mind, the company has promised to shield eligible customers from copyright claims linked to Imagen 2’s training data and outputs.
Image via Google
Imagen 2 is Google’s answer to the challenges faced by the first-gen model, which sometimes reproduced recognizable photos of real people or copyrighted artwork. This new version aims to navigate these murky waters with innovative training and modeling techniques, paving the way for a more responsible use of AI in image creation.
[via TechCrunch and Slashdot, images via Google]