Amazon has entered the AI image generation landscape with the Titan Image Generator, a text-to-image AI model presented at the AWS re:Invent conference. Aimed at an enterprise audience, this tool enables developers to build AI image creation applications with the capability to generate realistic, studio-quality images while addressing concerns related to toxicity and bias.
During the keynote at AWS re: Invent, Swami Sivasubramanian, AWS vice president of database, analytics, and machine learning, showcased the Titan Image Generator’s functionalities. In addition to creating images from natural language prompts, the model can change backgrounds, offering a versatile solution for developers.
Unlike standalone applications, Titan is not a separate platform but a model that developers can leverage by integrating it with Amazon Bedrock. All images generated using the Titan Image Generator will automatically include invisible watermarks, aligning with Amazon’s voluntary commitments made with the White House in July.
The invisible watermark serves as a means to mark images as AI-generated with Titan Image Generator, in accordance with the Biden administration’s executive order on AI. Vasi Philomin, vice president for generative AI at AWS, explained that the watermark is designed to be nonintrusive, with no impact on visual quality, latency, or susceptibility to cropping or compression. Notably, the watermark extends beyond metadata within the file.
To detect the invisible watermark, Amazon has introduced an API that users can connect to and use to verify an image’s provenance. This approach aligns with the model’s nature, as Titan is intended to be a tool for developers rather than an end product. Developers building with Titan Image Generator have the flexibility to decide how to convey information about the image’s AI-generated origin to users.
In addition to Titan Image Generator, Amazon announced the general availability of other Titan models, including Titan Text Lite, a smaller model for lighter text generation tasks, and Text Express, designed for more extensive use cases like conversational chat apps.
Furthermore, Amazon will extend copyright indemnity to customers utilizing its Titan foundation models, encompassing text-to-image functionality. Legal cover will also be provided to users of any Amazon-created AI application, even if the application utilizes a different foundation model from Amazon’s Bedrock AI model repository, such as Meta’s Llama 2 or Anthropic’s Claude 2. Some applications covered by this include AWS Health Scribe, CodeWhisperer, Amazon Personalise, Amazon Lex, and Amazon Connect Contact Lens.