Adobe has widely released a new and potentially contentious feature: text-to-image generation for Photoshop powered by Firefly, first teased in April. As with LLMs like Dall-E and Mid Journey, you can use it to create an image from scratch by typing a description into Photoshop’s updated generative AI tool.
I tried it with the text “Dramatic low angle view of a steamship from the 1800s in a storm with large waves and lightning” in multiple styles (anime, watercolor, sketch, realistic) and got decent results. The usual AI art caveats apply though, particularly with weird details if you look closely. But it certainly created useable results and you have the benefit of already being inside Photoshop to fix any errors.
Previously, Photoshop’s Generative Fill feature only let you add, extend or remove specific parts of an image. Now, you can create images from scratch, then tweak them later. “This really speeds up time to creation,” Adobe’s Erin Boyce told Engadget in April. “The idea of getting something from your mind to the canvas has never been easier.”
The feature is powered by Firefly Image 3 model, something at the heart of a recent artist backlash against Adobe. Creators were incensed by language in Adobe’s recent ToS (terms of service), interpreting it to mean that Adobe could freely use their work to train the company’s generative AI models.
In its latest post, however, Adobe stated that it has a “commitment to creator friendly AI” which means “never training on customer content.” It promised to take a creator-friendly approach as part of its AI ethics principles of accountability, responsibility and transparency.
Along with image generation, Adobe introduced an “Enhance Detail” feature in Photoshop’s Generative Fill. For Illustrator, it introduced Generative Shape Fill to add detailed vectors in a designer’s unique style (above), Enhanced Text to Pattern (creating customized vector patterns in the artists style) and Style Reference. It also added a Mockup tool to create “high-quality visual prototypes of art on objects like product packaging,” enhanced selection capabilities and more.