Adobe is finally joining the AI-generative game with the newly launched Adobe Firefly, an assortment of AI-generation models currently in beta. It will be eventually integrated into the Adobe ecosystem, primarily into Creative Cloud’s most popular apps, such as Photoshop, Illustrator, and Premiere Pro.
At this time, the company released one model that is a text-based generator for images and stylized text. But there are more developments to come.
The firm is also offering considerable transparency regarding the legal status of the tools, disclosing they’ve only been trained with out-of-copyright or approved content from Adobe Stock.
Let’s talk more about all this!
Adobe's new AI tool suite is the successor of the initial AI-generative features they introduced into Adobe Express and Photoshop during 2022’s Adobe Max conference. But it promises to be far more powerful and integrated across the Adobe board.
This time, they released the public beta –albeit available upon request only– of only one of what they announced will be a family of AI generative models. Adobe Firefly’s current tool is a text-to-image generator that produces pictures and stylized text with visual effects –some have described it as a modern WordArt–parting from text prompts introduced by the user. The images are intended for commercial use, and the results from the demos we’ve seen are really impressive.
In the future, they plan to release models that can generate vector variants, create 3D models or apply the look of a given image to an entire footage file, for example.
Right now, Adobe Firefly’s text-to-image generator is available on its own web page and upon request. But eventually, all the models will be readily available within Adobe Express, Photoshop, Illustrator, Premiere Pro, and even across Document Cloud and Experience Cloud.
Firefly’s functionality right now is similar to today's popular AI image generators, like Dall-E, Stable Diffusion, or Midjourney.
However, Adobe says their model focuses more on user-friendliness by eliminating the need for complex and extremely detailed text prompts (although you can still use it that way if you wish).
Firefly comes with built-in, filter-like presets for things like visual style, lighting, or aspect ratio, which helps you realize your vision without too much text description. The preset features are also useful for Firefly’s other functionality, AI image editing: applying a certain style to existing images.
Simplifying the creation process is something a competitor firm is also focusing on. Shutterstock Generate, the stock photo agency’s native AI image generator built on top of Dall-E technology, also aims at producing images from simpler text prompts.
It seems these firms understand that democratizing access to AI models’ best results for all users (not just text-prompt-skilled users) is the way to go.
Adobe did not shy away from the pressing issue surrounding the legality of AI-generated media and disclosed a whole plan for transparency and ethical management of the models.
For one, the company assures all their models have been trained only with out-of-copyright content and licensable content from Adobe Stock –their own stock media marketplace– of which they have the rights cleared for this use. This sets them aside from other tools like Stable Diffusion and Midjourney, which are currently under legal fire for training their software with copyrighted content from the web.
By not feeding other artists’ work or trademarked brands’ data into their software, the firm can ensure that Firefly isn’t able to replicate other people’s intellectual property, another point of contention for AI-generated images that have proven to exactly replicate copyrighted photos, logos, and even watermarks.
Finally, Adobe also shared that they are planning a scheme to reward Adobe Stock artists for the use of their work in Firefly training, although it’s not yet disclosed how this will be done (it could be similar to Shutterstock’s Contributor Fund for this specific end). Equally relevant, they said they will implement an opt-out system for artists to label their work as Do Not Train within metadata, clearly establishing their refusal to participate in AI training datasets.
The Do Not Train tag is part of the Content Authenticity Initiative (CAI) by the Coalition for Content Provenance and Authenticity, two initiatives the company impulses to establish new industry standards around AI-generated media.
The one thing worth noting here is that the firm cannot, at this time, guarantee whether Firefly-generated content would be copyrightable, especially after the latest guidance on the copyright of AI-generated media by the USCO. They did say they believe their models give way for humans to have enough creative control for copyright to be a possibility.
While Adobe took a long in formally introduce itself into the new AI-generative models' space, they seem to have effectively leveraged its decades of experience in creative software and serving the creative community, with models that have taken every possible need into consideration and tried to solve every relevant issue before it arose.
The initial results shared from Adobe Firefly are rather impressive, and if user-friendliness is as real as it seems, then it’ll be no doubt a valuable tool to add to their Creative Cloud suite.
Are you signing up to try Adobe Firefly? Let us know how you find it!
I am an experienced author with expertise in digital communication, stock media, design, and creative tools. I have closely followed and reported on AI developments in this field since its early days. I have gained valuable industry insight through my work with leading digital media professionals since 2014.
AI Secrets is a platform for tech decision-makers to learn about AI technology. Our team includes experts such as Amos Struck (20+ yrs ICT, Stock Photo, AI), Ivanna Attie (expert in digital comms, design, stock media), and more who share their views on AI.
Get AI news in your inbox & join thousands of engineers and managers using AI to boost sales and grow market share