Adobe Firefly is an AI-powered tool that is used to create digital effects from text prompts. The Text Effect tool is still in beta, but users have already started creating amazing images using just key words or phrases.
While early use examples are impressive, what we’re seeing now in generative AI digital arts and photography is a snapshot in time and technology—that is where the tool is today. While artificial intelligence has been decades in the making, the technology is still in the learning and development phase. As good as it is now, we will continue to see improvements and refinements in the coming months that will render AI generated images more precisely.
User feedback is the real powerhouse as those ‘thumbs ups’ and ‘thumbs downs’ communicate to the AI tool how close to the user’s expectation the generated creations are. This engagement between users and machines helps the tool become ‘smarter’ as it continuously hones its output response. For example, today if you ask an AI tool to create a frog image, it will dip into its data set and produce a green or brown frog—something that a human has already confirmed to be accurate. If you want a blue frog, you’ll have to specify ‘blue’ as well as ‘frog’ because the AI tool only has green or brown frog images as references. The tool will not be able to create blue frogs without the additional text prompt of ‘blue’ to output the expected frog. In theory, as blue frogs are added to the data set, and if a significant number of users always reject the green or brown frog, and only accept the blue frogs that had been previously requested, the AI will eventually figure out that ‘blue frog’ is what is meant when the keyword ‘frog’ is input into the AI tool, and only blue frogs will be generated. The AI tool depends on a catalog of images in its data set. Too few reference images, and the tool does its best with what little it has, but will likely fall short of expectations. It also depends on user imput for training. This example created by Adobe Firefly (Beta) Text Effects, the prompt was “Rubik’s Cube”, a toy invented in 1974 and wildly popular throughout the 1980’s. We should expect a construction of saturated red, blue, yellow, orange, green and white cubes.
While the output is interesting, it bears no resemblance to a Rubik’s Cube. We have some cube shapes, but the color palette is more pastel and for most people, they would not guess the prompt was “Rubiks’ Cube”. The AI tool apparently needs more training regarding Rubik’s Cubes in its data set to recreate it adequately. The user can communicate via a thumbs up or down, how closely the output matches the expectation. In this case, users would reject the output and train the AI tool toward a more acceptable colored design. We will check back in a few months and compare.
In contrast, the prompt for this design was “red electric guitar”. Here the AI tool has been trained to pull out recognizable red electric guitar components. You can see voids in some letter shapes where the AI wasn’t able fill in the shapes exactly, but over-all it did a very good job.
Try it out for yourself
Right now Adobe Firefly is free in its beta form to be used in a browser. Go to firefly.adobe.com and send a request to be a part of the group. It may take some time to be accepted, but once you are, you’ll be able to test all the upcoming generative AI tools on the horizon, including vector files and videos as their beta versions are released.
We’ve experimented with the Adobe Firefly Text Effects tool and came up with 26 prompts to get you started. Because the tool is always learning every minute of every day, it’s not possible to recreate these prompts exactly as we’ve done here. Your creations using our prompts will likely be more refined. See our website catalog for all our prompts.
Disclaimer. All images are samples from the Adobe Firefly (Beta) and are not available for commercial use at this time. The images in this blog and catalog are for informative purposes only to promote Adobe Firefly and to provide examples of generative AI.
The prompt for our blog title image is: manhole cover stained glass