Amazon announced the launch of several new artificial intelligence services at its re:Invent conference today, including an image generation model called Titan Image Generator. With the release, Amazon joins fellow tech giants like Google, Meta, and Microsoft in releasing AI systems that can create original images and art from text prompts.

Titan Image Generator is now part of Amazon’s Titan suite of AI services available through Bedrock on Amazon Web Services (AWS), the company’s cloud computing division. Users can provide text prompts to have Titan Image Generator produce original images. The system can also edit existing images by changing or removing backgrounds.

Swami Sivasubramanian, Amazon Web Services’ (AWS) vice president of database, analytics, and machine learning, framed the image modification feature as a major selling point during his keynote speech. He emphasized how the tool enables the creation of lifestyle images while retaining the main subject of the image, underscoring the feature’s potential to revolutionize visual content generation across a wide range of industries.

Image generation built for enterprises

Amazon has chosen to release Titan Image Generator not as a standalone app or service, but instead, designed it as a developer tool to be used in building applications powered by the model. This approach is aimed squarely at an enterprise audience, differentiating Amazon’s offering from more consumer-focused image generators such as OpenAI’s DALL-E.

The launch of Titan Image Generator comes amid increasing scrutiny around AI art generators and their potential to replicate copyrighted images or create harmful content. Amazon says Titan Image Generator has built-in safeguards against biases and includes invisible watermarks on all images to identify them as AI-generated.

It’s important to note, however, that Amazon’s watermarking approach appears to be its own proprietary solution, diverging from other tech companies that have embraced the Content Credentials system developed by the Coalition for Content Provenance and Authenticity (C2PA). This raises questions about interoperability and the broader ecosystem of tools for detecting and validating these watermarks.

New models, legal safeguards, and a bold strategy

In addition to the Titan Image Generator, Amazon also announced the availability of other Titan models. These include Titan Text Lite, a smaller model used for lighter text generation tasks like copywriting, and Text Express, designed for larger tasks such as powering conversational chat apps.

Perhaps most notably, Amazon has extended copyright indemnity to customers who use its Titan foundation models, including the text-to-image model.

This legal cover applies even if users employ a different foundation model available in Amazon’s Bedrock AI model repository, like Meta’s Llama 2 or Anthropic’s Claude 2. This move is a significant reassurance for AWS customers worried about potential copyright issues when using generative AI.

Amazon’s foray into AI image generation signifies a new phase in the company’s AI strategy. By offering powerful tools and legal safeguards, Amazon is poised to make a significant impact on the landscape of AI image generation and use. As the technology continues to evolve, the Titan Image Generator may prove to be a catalyst for novel applications in AI-driven content creation.