Today marks one year since Adobe introduced Firefly, the logical evolution of the already extensive set of AI-based features that fueled its suite of tools and services. It is true that, thanks to the big boom of chatbots during 2023, the general public associates generative models mainly with text tools, but in fact Adobe has been applying them to audiovisual content for a long time and with excellent results.

Since, we experienced the arrival of new features, like those that initially landed in the premiere, later some that left us speechless in Photoshop, and recently, last October, all the new features and technologies that were tested, that were shown to us in Adobe Max and that we compiled here and here. And to extend the reach even further, Firefly recently finally arrived in Adobe Express versions for Android and iOS.

This is something I think about from time to time, but it seems quite important to me, and although there are other technologies today that we usually associate with the development of artificial intelligence, Adobe has been using it for many years in a very innovative way and which is also extremely practical for users of its tools. Not to mention testing technologies that were years ahead of other solutions that hit the market, such as what we saw with Adobe VoCo at the time.

Now, to celebrate the birthday of unifying our AI-based solutions, Adobe Firefly adds reference structure functionalitywhich greatly increases the control the user has when using the image generation tool behind call text. and this, as a general rule, requires particularly comprehensive descriptions if we are clear about what we want to gain in the process.

An example of using a structure reference. On the left is the original image and on the right four generations of Adobe Firefly starting from it and modifying the style.

As we’ve seen in other AI-based image generation tools, Adobe Firefly also allows you to use a reference image in combination with call of text, but in this particular case and with the advent of the structure reference, which can be combined with the style reference, The model allows the user to reference both the structure and style of the imageto mark specific elements we want to be specific to the new image later in the same prompt.

One of the most common problems that users face when using an AI imaging tool is that in many cases, even if they specify very specific parameters of its content, the models do not correctly derive what we want, and as a result, give us a result that which is very far from what was desired. With this new Adobe feature and if we already have a reference image, the process will be much faster, easier and more reliable. As an example, Adobe gives three practical examples of its use:

• Change the design of an entire room with the push of a button: Upload a picture or sketch of an existing room and press “Generate”.
• Turn a child’s drawing into a great picture or change the style of the picture: from painterly to photorealistic or vice versa. Really bring your drawings to life by making them photorealistic with added depth, texture, lighting, color and more.
• Colorize black and white images and line drawings to create a completely new theme that has nothing to do with the reference image and maintains the same structure.

More information