Amazon has enhanced its in-house image-generating model, Titan Image Generator, for AWS consumers who utilize its Bedrock generative AI platform
According to a blog post by AWS principal developer advocate Channy Yun, the new model, which is referred to as Titan Image Generator v2, introduces several new capabilities. According to Yun, users can “guide” the images they create by utilizing reference images, editing existing visuals, removing backgrounds, and generating variations of images.
According to Yun, the Titan Image Generator v2 can intelligently detect and segment numerous foreground objects. “The Titan Image Generator v2 enables the creation of color-conditioned images that are predetermined by a color palette.” [And] the image conditioning feature can mold your creations.
Titan Image Generator v2 supports image conditioning, which involves the optional inclusion of a reference image and the emphasis on specific visual characteristics in that image, such as edges, object outlines, and structural elements. Reference images, such as a product or company logo, can also refine the model, ensuring that the generated images maintain a consistent aesthetic.
AWS maintains a lack of clarity regarding the specific data it employs to train its Titan Image Generator models. The company had previously disclosed to TechCrunch that the data is a blend of proprietary and licensed information.
Few vendors are willing to disclose this information; they regard training data as a competitive advantage and, as a result, maintain it and the information associated with it in the strictest confidence. Training data details are also a potential source of IP-related litigation, which is another disincentive to disclose too much.
Instead of transparency, AWS provides an indemnification policy that protects customers if a Titan model, such as Titan Image Generator v2, regurgitates (i.e., spits out a mirror replica of) a potentially copyrighted training example.
In the most recent second-quarter earnings call, Amazon CEO Andy Jassy expressed his continued “very bullish” sentiment regarding generative AI technology, such as AWS’ Titan models, despite the enterprise’s tendency to second-guess the technology and the increasing expenses associated with training, fine-tuning, and serving models.
“The generative AI space is poised for rapid growth,” he stated, “and it will be primarily constructed in the cloud from the outset.”
