November 29th – As many Cyber Talk readers know, Amazon organizes the annual AWS re:Invent conference, where it makes announcements, launches exciting new products, unveils new product features and releases new tools.
The premiere event is currently in underway in Las Vegas. Early this morning, Amazon debuted its new Titan Image Generator, which is now available in preview mode for AWS customers.
Titan Image Generator
Part of Amazon’s Titan family of generative AI models, the image generator can be prompted with text to create new images, or it can customize existing images. Says Amazon, the Titan Image Generator was trained on diverse sets of data, and includes mitigations for toxicity and bias.
However, the company declined to explain where the data sets came from, whether it obtained permission from creators to use said images, and whether or not it’s compensating all creators of the images that were deployed in training the AI models.
Watermark by default
In an effort to mitigate the proliferation of AI-based misinformation, images created with the Titan Image Generator will have a “tamper-resistant” invisible watermark on them, by default. Nonetheless, customers may be able to remove the watermarking.
The ability to identify content as AI-generated is a core element of the White House’s executive order on AI.
Examples of capabilities
- A sample prompt for a new image might read as: Please produce an image featuring [our product] in the context of [specific situation] and portray it in a way that will stand out to consumers.
- A sample prompt for customization of an existing image might read as: Please swap out the background of this image with a new background that depicts [specific type of scenery].
More Amazon AI models
At the event, Amazon also announced the general availability of other Titan models. These include Titan Text Lite, which is a smaller model intended for text generation tasks, like copywriting. Another model known as Text Express is intended for conversational chat apps.
Copyright indemnity will be extended to customers who use its Titan foundation models, including text-to-image. Legal cover will also be offered to those who use any Amazon-created AI application, even if the app leveraged a different foundation model from its Bedrock AI model repository (ex. Meta’s Llama 2 or Anthropic’s Claude 2). Such apps include AWS’s HealthScribe, CodeWhisperer, Amazon Personalize, Amazon Lex, and Amazon Connect Contact Lens.
For more insights into AWS re:Invent and the cyber space, please see this Check Point blog article. Lastly, to receive timely cyber security insights and cutting-edge analyses, please sign up for the cybertalk.org newsletter.