Generative AI: Adobe assures that Firefly corrects the disadvantages of DALL-E

Image generators using generative artificial intelligence are now famous thanks in particular to DALL-E or Bing Image Creator. Since then, many competitors, including Adobe’s Firefly, have entered the scene. And now Adobe is launching major updates to the features and availability of Firefly.

Adobe announced Tuesday the commercialization of Firefly for Generative AI, after a six-month beta period. With Adobe Firefly, users can use text prompt prompts in more than 100 languages to generate images, text effects, and vectors.

“With more than 2 billion generations, the creators have amazed us with their commitment and feedback on the Firefly beta, which prompted us to offer generative AI features designed to be commercially secure and seamlessly integrated into the interfaces appreciated by customers,” said Ely Greenfield, chief technology officer of Adobe.

The Firefly model is based on Adobe Stock images and public domain content whose copyrights have expired

One of the main differentiating factors between Firefly and other AI image generators, such as DALL-E, is that the Firefly model is trained on Adobe Stock images and public domain content whose copyrights have expired, which ensures that the content used to train the models was done with the creator’s permission.

In addition, Adobe Stock contributors will receive compensation for any content used to train the commercial Firefly AI model. Adobe describes this mechanic as a way to establish a fair partnership with image creators. The references of the content will be included in all the generated images to serve as a “nutritional label” with verifiable details that show the name of an asset, the date of creation, the tools used for the creation and the modifications made.

All these measures respond to the main controversies raised by AI image generators, in particular the dissemination of false information through realistic images (deep fake) and the misappropriation of the work of others to train the models.

Testing on the web

Adobe has also unveiled a new Firefly web application that allows users to test the features for themselves – what the company describes as “a playground for exploring AI-assisted creative expression”. Users just need to visit the web application, log in or create an Adobe account, and then tinker with as many AI models as they want.

Users can also access the features supported by Firefly in Adobe Creative Cloud and Adove Express, including Generative Fill and Generative Expand in Photoshop, Generative Recolor in Illustrator, and Text to Image and Text effects in Adobe Express, according to the press release.

Finally, Adobe has introduced Generative Credits, which give paying customers of the Firefly web application, Express Premium and Creative Cloud a token allocation for generative AI image workflows. Once the specific generative credits have been exhausted, subscribers have the option to continue using the generative AI features with a slower experience or to purchase additional subscription packs.

Source: “ZDNet.com “