Amazon has released an updated version of its internal image generation model, Titan Image Generator, for AWS customers using its generative AI platform Bedrock.
The recent model, simply called Titan Image Generator v2, brings several recent features, explains Channy Yun, Principal Developer Advocate at AWS in a blog postUsers can “guide” the pictures they generate using reference images, edit existing graphics, remove backgrounds and generate image variations, Yun says.
“Titan Image Generator v2 can intelligently detect and segment multiple foreground objects,” Yun writes. “With Titan Image Generator v2, you may generate color-conditioned images based on a color palette. (And) you need to use the image conditioning feature to provide shape to your creations.”
Titan Image Generator v2 supports image preparation by optionally using a reference image and specializing in specific visual features in that image akin to edges, object outlines, and structural elements. The model can be fine-tuned using reference images akin to a product or company logo in order that the generated images maintain a consistent aesthetic.
AWS stays vague about exactly what data it uses to coach its Titan Image Generator models, previously telling TechCrunch only that it’s a mix of proprietary and licensed data.
Few vendors willingly disclose such information. They see training data as a competitive advantage and subsequently keep it and the knowledge related to it strictly confidential. Details of coaching data are also a possible source of IP-related litigation, which is another excuse not to reveal an excessive amount of.
Instead of transparency, AWS offers a compensation policy that protects customers within the event that a Titan model akin to Titan Image Generator v2 replays (i.e., spits out a mirrored copy of) a potentially copyrighted training example.
In the corporate's recent second-quarter earnings call, Amazon CEO Andy Jassy said he was still “very optimistic” about generative AI technology like AWS' Titan models, despite signs of doubt from the corporate and the rising costs related to training, tuning and deploying models.
“The field of generative AI goes to get big quickly,” he said, “and for essentially the most part, every little thing shall be inbuilt the cloud from the beginning.”
