Firefly 3: AI updates generate excitement at Adobe MAX London

Adobe MAX came to London this week, revealing the third iteration of the Firefly Image Model to the world and unveiling a host of AI enhancements to the Creative Cloud graphics applications.

First to take the stage at Battersea Evolution was Ashley Still, General Manager Adobe Creative Cloud, who introduced the latest iteration of Firefly Image Model, “a major advancement in quality control and consistency”, available now as a beta on adobe.com.

“We believe creativity is a uniquely human trait,” she stated. “We view our role as helping you navigate and thrive through this change by unlocking the power of AI as a tool for you”.

Still said that in the year since the release of Firefly, the “response has been phenomenal”. “You have generated over 7 billion images with Firefly,” she said. “You’ve told us that we got some important things right like the integration of Firefly powered features like Generative Fill in Photoshop and Generative Recolour in illustrator to give you the creative control that you want increasing your productivity without limiting your creativity.”

“[With Firefly Image 3 Model] we’ve significantly improved quality, with photorealistic results, fine details, better lighting, and positioning,” she said. “We’ve developed new stylisation techniques that give you a wider variety of results to choose from. When generating people, improved quality means detailed features and just more personality with a broader range of moods and expressions.”

“Improved prompt adherence” means Firefly more accurately generates content from text prompts, while features like Structure Reference and Style Reference enable users to direct Firefly to create content based on their own reference images.

“You don’t have to rely on crafting the perfect prompt,” said Still. “With Image 3 Model you can ideate, explore, and create content faster and more fluidly than before. And like the rest of the Firefly family of models. It’s designed to be commercially safe, so you can use it with confidence in your production work.”

AI in Photoshop

With the launch of the latest Photoshop beta, Firefly is now integrated within the venerable image editor. It offers Generate Image, Generate Background, Reference Image, Generate Similar, and Enhance Detail.

At an event the evening before MAX, Terry White, Principal Worldwide Design & Photography Evangelist for Adobe, showed me how generative removal of objects in Photoshop and font replacement have both become more streamlined. To demonstrate Generate Background, he placed his dog Pixel on a selection of beaches completely generated from a text prompt, all complete with accurate shadows and realistic orientation. “It’s not only matching the perspective, the angle, the lighting, the shadows, it’s putting Pixel’s paws in the sand,” said White. “It’s doing all that compositing work for me that I would have had to do manually, and it did that in a matter of 15 seconds.”

White also showed me how a stock photo of a jacket can be used as a Reference Image and applied it to himself in a photo, with Firefly ‘fitting’ the jacket perfectly, complete with folds and shadows.

Lightroom isn’t missing out. As well as the AI driven Lens Blur feature, a forthcoming Firefly treat is a mask and generative replace feature in Lightroom, bringing it into line with Photoshop; it also works on Lightroom Mobile.

At MAX, White was on stage to reveal all this magic to the world, including a tool for identifying and selecting people in a shot and generating content to remove them. It was fast and impressive, at least when the ‘demo gods’ were playing nice. Firefly uses the cloud to compute and generate its content, so bandwidth is a key factor. However rest assured, it worked very smoothly when White demoed the Firefly features to me on his laptop and phone. Mac users take note – Mac OS 13.4.0 or later is recommended for the Photoshop Beta with Firefly, but I’ve found it can run fine on Mac OS Monterey.

Following White on the MAX main stage, Sr. Digital Imaging Evangelist & Community Advocate Anna McNaught demonstrated Firefly features within one of her signature compositions. “Generate Image is perfect for brainstorming ideas or for coming up with different backgrounds or creating skies,” she said. “I love using Generative Fill specifically for blending because it looks at two images to see how it needs to blend them both together. If I need to give myself a little bit more working room, I can use Generative Expand [the feature was released last year] to easily do that.”

Photoshop also features an Adjustment Brush tool that lets you paint adjustments on one part of an image, while aother enhancement is the ability to export adjustment presets. “I use adjustment layers all the time” said McNaught. “And I love that now I can package them up and apply this to every single piece of work that I create. And I think it’s super cool that you can share that with everyone as well.”

As well as demoing the existing Generative Recolour feature in Adobe Illustrator which applied new colours to a vector rocket object made up of 1000 shapes in a few seconds. Michael Fugoso, Senior Design Evangelist for Next Gen Creative Pros, demonstrated a very impressive generative expand feature for vector images. “This is going to be such a game changer of a tool for me,” he said. There was also MockUp, for previewing how illustrations look on physical objects, and a new Dimension tool.

Fugoso also demoed Firefly Text to Image built into InDesign, as well as a feature to generate images within text.

Adobe Express has also been dabbed with the Firefly wand, bringing Generative Fill and Text to Template to the social media content creation app.

Adobe is also developing AI-powered video editing capabilities, including a new Firefly video model and content credentials integration into Premiere Pro. Apart from the following teaser video shown on the main stage, nothing more was announced at MAX.