Amazon is set to unveil a new tool in the form of an image generator, following the trend of numerous tech giants and startups that have already ventured into this domain.
Unveiled during a keynote at the re:Invent conference, Amazon introduced the Titan Image Generator, now accessible in preview for AWS users. As part of Amazon’s Titan generative AI models, the Titan Image Generator has the capability to generate fresh images based on textual descriptions or modify existing images.
According to Swami Sivasubramanian, users can leverage the model to effortlessly change an image’s background to, for instance, a rainforest setting. The tool also allows for seamless background swaps to produce lifestyle images while preserving the main subject within the image and offering additional customization options.
Amazon highlights that the Titan Image Generator underwent training on a varied set of datasets spanning diverse domains, with the option for further refinement on personalized datasets. The tool incorporates built-in safeguards against toxicity and bias. However, Amazon has not disclosed the exact sources of these datasets or clarified whether consent was obtained from all image creators utilized in the model’s training process.
Sivasubramanian assured the audience that Amazon will defend customers facing copyright infringement allegations related to images generated using the Titan Image Generator, aligning with its AI indemnification policy.
Emphasizing the meticulous approach to model training and data selection, Sivasubramanian stated that Amazon’s customers will have the freedom to choose optimal models for their evolving needs.
Moreover, images produced through the Titan Image Generator will include an inherent “tamper-resistant” invisible watermark by default to combat the proliferation of AI-generated misinformation, although users likely have the option to disable this feature.
