Amazon has joined the race of image generation using artificial intelligence by unveiling the Titan model to convert text into images.
This was announced during the re:Invent conference, where the Titan image generator can create high-quality, realistic images. It is supposed to include safeguards against bias and discrimination.
Titan is not a standalone application or website, but a tool that developers can rely on to build image generators based on the model. Developers need access to Amazon Bedrock to use it.
During his keynote speech, Rajeev Sivasubramanian, Vice President of AWS for databases, analytics, and machine learning, reviewed the Titan Image Generator device and pointed out that the model is capable of creating natural language variable images and also changing backgrounds.
The main target of the Titan image generator is to directly target enterprise audiences, rather than focusing heavily on the average user like current image generation models based on artificial intelligence do.
All images produced by the Titan Image Generator automatically contain invisible watermarks as described by Amazon.
Vasili Filomen, Vice President of Generative AI at Amazon Web Services, mentioned that this is part of Amazon’s voluntary commitments signed in July with the White House.
Filomen said, “We were looking for a way to distinguish the image as being generated by artificial intelligence, especially using the Titan image generator, without affecting the visual image itself, without affecting its transmission time, not able to be cropped or compressed”, adding that the watermark is not limited to file’s descriptive data.
Amazon has created an API that individuals can use to examine the image source. Amazon enables developers using the Titan Image Generator to choose how to display this information to users.
The watermark is an essential part of the Biden administration’s AI management plan. Companies like Adobe and Microsoft have adopted the “Content Credentials” system developed by the C2PA alliance to address this issue.
