Use Firefly’s Generative Match to generate images in your style

Generative Match, launched in beta, is a product of Adobe’s responsible AI development

The ability to conjure up an original image just by entering into Adobe Firefly’s prompt box a few words — like “a bear in the woods eating an ice cream cone” — never ceases to amaze me. But text prompts alone don’t suffice. What’s harder to write down is exactly how you want something to look — finding the exact words to describe the style you have in your mind’s eye can be nearly impossible.

That’s why we are launching a new Firefly capability, available in beta, called Generative Match. With Generative Match, you can include a reference image along with your text prompt. Firefly’s generative AI will produce imagery that combines both your text prompt and your reference image. If your reference image shows a cat in a whimsical cartoon style, the bear in your new image will sport the same look. If the background in your reference image includes lots of purples, browns, and reds, the woods in your new image will, too.

Generative Match is a powerful technology that will help creators leverage their work — or their brand’s style — to make new images at scale that share a consistent style and feel. We’re excited to see what you make with Generative Match. But we also realize that this feature, like all generative AI technology, must be developed responsibly with strong safeguards against improper use — and we’re doing that.

To learn about more new capabilities coming to Adobe Firefly, check out this post.

How Generative Match works

Start in the Firefly web application, our site for AI-assisted creativity. When you enter your text prompt in the Text-to-Image tool, look for “Style Match” on the right. You’ll be able to either pick a reference image from a pre-selected set of images in various styles, licensed by Adobe, or upload an image of your own — provided you confirm that you have the rights to use the image.

And as with all Firefly features, you’re in charge. Maybe you want your generated image to look just like your reference image. Or maybe you want something with more of its own character. With a simple slider, you can determine how close the resemblance is between your reference image and what you generate.

We’ve implemented Generative Match technology in the Firefly website and Adobe Illustrator first, then we’ll roll it out to other Creative Cloud tools.

There are a lot of benefits to generating an image with Generative Match. Most artists have a style that expresses their unique perspective. With Generative Match, you can guide Firefly to produce images in one of your own unique styles. You’ll almost certainly take those images into a tool like Adobe Photoshop so you can tweak them and make them your own, but Generative Match gives you a head start.

If you’re on a marketing team working on a new campaign, Generative Match will help your team produce campaign content with a consistent look, even if there are many people working on assets.

We’re excited because Generative Match, like Firefly more broadly, can help creative people keep up with the tremendous demand for content. It will also make exploring different creative approaches much quicker and easier — and many of you have told us: the more cycles of exploration you have and the more possibilities you encounter, the better solutions you create.

Responsibility and protections

As we’ve developed Firefly to empower creative careers, we have at the same time pursued a parallel track of innovation to prohibit abuse, discourage bad actors, and encourage transparency and attribution.

We’ve trained our Firefly generative models responsibly on licensed content and public domain content for which copyright has expired. And we apply content moderation that filters assets before they become part of the Firefly training dataset. In addition to Firefly being designed to be commercially safe, Adobe is also offering enterprise customers the opportunity to obtain IP indemnification for Firefly-generated content.

We led development of Content Credentials, a kind of “digital nutrition label” that allows creators to attach to their work information such as their name, the date, and the tools and edits used to create a piece of content. This information is designed to bring more transparency and trust to digital content. This open-standard approach was driven by the Content Authenticity Initiative, which was co-founded by Adobe and now includes nearly 2,000 members. Major brands such as Microsoft, Leica Camera, Nikon, Publicis Groupe and many others have integrated Content Credentials into their digital experiences. (Learn more about how these companies are deploying Content Credentials at enterprise scale and across major product lines here.)

We have also helped draft, and are advocating that Congress pass, the FAIR Act, which will give creators the right to take action against people misusing AI to intentionally impersonate their style or likeness for commercial gain without their permission.

And we’ve developed new policies and safeguards to promote legitimate use of new features like Generative Match and discourage bad actors. First, we require users to confirm they have the right to use any work that they upload to Generative Match as a reference image.

Second, if an image’s Content Credentials include tags indicating that the image shouldn’t be used as a style reference, users won’t be able to use it with Generative Match. We will be rolling out the ability to add these tags to assets as part of the Content Credentials framework within our flagship products.

Third, when a reference image is used to generate an asset, we save a thumbnail of the image to help ensure that the use of Generative Match meets our terms of service. We also note that a reference image was used in the asset’s Content Credentials. Storing the reference image provides an important dose of accountability. Our Firefly indemnification will not cover claims over assets that are based on infringing content provided by a user. (Note: Stored reference images are not used to train Firefly.)

As a reminder, Generative Match is a new feature and will remain in beta as we get your feedback and iterate both the functionality and the safeguards we have in place. We are developing all Firefly features in collaboration with the creative community. We want to hear from you about what’s working and what we can improve, so please join the conversation on Discord or the Adobe Firefly Community forum to exchange ideas, share your creations, and provide feedback.