Amazon has released an upgraded version of its home-grown image generation model, Titan Image Generator, for AWS customers using its Bedrock generative AI platform.
The new model, simply called Titan Image Generator v2, packs several new features, as Channy Yun, principal developer advocate at AWS, explains in a blog post. Yun says that users can use reference images to “guide” the image they generate, edit existing images, remove backgrounds, and generate variations of an image.
“Titan Image Generator v2 can intelligently detect and segment multiple foreground objects,” Yun wrote. “Titan Image Generator v2 allows you to generate color-adjusted images based on your color palette. [And] Use the image adjustment features to help shape your creation.”
Image credit: Amazon
Titan Image Generator v2 supports image refinement, optionally incorporating a reference image to focus on specific visual characteristics in the image, such as edges, object outlines, structural elements, etc. Additionally, reference images such as product or company logos can be used to fine-tune the model, ensuring consistent aesthetics in the generated images.
Image credit: Amazon
AWS remains unclear about what data it uses to train its Titan Image Generator models, with the company previously telling TechCrunch only that it's a combination of its own and licensed data.
Few vendors are willing to share this information publicly; they consider training data a competitive advantage, so training data and related information is kept a closely held secret. Details of training data are also a potential source of IP-related litigation, which also serves as a deterrent to releasing much information publicly.
In exchange for transparency, AWS offers an indemnification policy that protects customers if a Titan model, such as Titan Image Generator v2, reproduces (i.e. spits out mirror copies of) training examples that may be copyrighted.
During the company's recent second-quarter earnings call, Amazon CEO Andy Jassy said he remains “very bullish” on generative AI technologies like AWS' Titan models, despite signs of skepticism from businesses and the growing costs associated with training, fine-tuning and serving the models.
“Generative AI is going to get big very quickly,” he says, “and most of it will be built in the cloud from the ground up.”