Amazon has entered the realm of AI image generation by introducing its Titan text-to-image AI model. Unveiled at the AWS re:Invent conference, the Titan Image Generator boasts the capability to produce “realistic, studio-quality images” while incorporating safeguards against toxicity and bias. Unlike a standalone platform, Titan serves as a tool for developers to construct their image generators utilizing the model, requiring access to Amazon Bedrock.
During his keynote at the conference, Swami Sivasubramanian, AWS’s vice president of database, analytics, and machine learning, showcased the Titan Image Generator’s features, highlighting its capacity not only to generate images from natural language prompts but also to alter backgrounds. Targeting an enterprise audience, Titan distinguishes itself from consumer-centric image generators such as OpenAI’s DALL-E.
All images produced by the Titan Image Generator will come with imperceptible watermarks, as stated by Amazon. Vasi Philomin, AWS’s vice president for generative AI, explained to The Verge that this measure aligns with the commitments Amazon made with the White House in July.
Philomin emphasized that the watermark is not limited to the file’s metadata and cannot be removed through cropping or compression. To detect these watermarks, Amazon developed an API for users to verify the image’s origin, underscoring the intentional separation of Titan as a model from a finished product. Developers leveraging Titan Image Generator have the flexibility to determine how to disclose this information to end-users.
The concept of watermarking images to denote AI-generated content is a pivotal aspect of the Biden administration’s AI executive order. In response, companies like Microsoft and Adobe have embraced the Content Credentials system established by the Coalition for Content Provenance and Authenticity (C2PA). Adobe has gone a step further by introducing an icon to signify content credentials in image and video content.
In addition to Titan Image Generator, Sivasubramanian announced the availability of other Titan models: Titan Text Lite, designed for lighter text generation tasks such as copywriting, and Text Express, tailored for more extensive applications like conversational chat apps.
Amazon is extending copyright protection to customers utilizing its Titan foundation models, including text-to-image capabilities. Furthermore, legal coverage will be provided to users of any Amazon-developed AI application, even if the application is based on a different foundation model from the Bedrock AI model repository, such as Meta’s Llama 2 or Anthropic’s Claude 2. Notable applications encompassed by this initiative include AWS HealthScribe, CodeWhisperer, Amazon Personalize, Amazon Lex, and Amazon Connect Contact Lens.
