Adobe Launches Firefly Generative AI Models for Synthetic Images and Font Design
Adobe introduced a new set of generative AI models this week called Firefly. The company has long-term ambitions for spreading Friefly’s creative potential, starting with a text-to-image generator and a tool for adding and manipulating the look and style of words in a text.
Adobe Firefly’s text-to-image engine performs much like OpenAI’s DALL-E or the Stability AI-based Midjourney. To stand out, Adobe included additional menus to set the size, lighting, and art style, rather than relying on the user to think up all of the right terms and remember every component. The effects can also be used to revamp an existing image. The text effects have a similar setup and range of options, and the AI relies on a mix of AI models to process and produce visuals. Firefly is technically in beta and is only accessible through a website right now. Adobe’s vision for the technology involves embedding an ever-growing collection of generative AI features in Photoshop, Illustrator, Premiere, and the rest of its creative software portfolio. The features will be priced into those services, not as stand-alone products for purchase.
“Generative AI is the next evolution of AI-driven creativity and productivity, transforming the conversation between creator and computer into something more natural, intuitive and powerful,” Adobe president of digital media business David Wadhwani said. “With Firefly, Adobe will bring generative AI-powered ‘creative ingredients’ directly into customers’ workflows, increasing productivity and creative expression for all creators from high-end creative professionals to the long tail of the creator economy.”
Notably, Adobe has disclosed the sources used to train its models, highlighting that all of the data is either part of its own library, specifically licensed for training, or in the public domain. That way, Adobe can claim it never violates any trademark or other intellectual property rights. The company also said it is working on a payment system for those who own any work it wants to use for its AI training program. The company is further committing to that standard with the development of a “Do Not Train” system that will let artists incorporate a request into an image’s metadata to block data scrapers from using it to train AI.
“We believe that Firefly can help people grow their creative confidence by removing barriers between creative vision and blank page,” Wadhwani wrote in a separate blog post. “It puts hours back into the lives of creators at a moment when both the demand for content and the demands on their time have exploded. We want Firefly to realize the power of generative AI and amplify human creativity and intelligence without replacing the beauty and power of human imagination.”