How to Prompt in Adobe Firefly

With a feature-rich UI and a unique training system, prompting in Adobe’s new AI image generator, Firefly, looks a lot different than the other leading generative AIs.

Many of the usual propting tricks you used in Midjourney and Dall-e will not work in Firefly.

This guide runs you through all the ins and outs of prompting in Firefly.

Use these quick links to quickly jump through out the past.

Related Articles

• Firefly V3
• Illustrator Text-to-Vector

Quick Links

1. Don’t Use Artist Names
2. Use Adobe Stock for Inspiration
3. Browse the Firefly Gallery
4. Use the Sidebar
5. Use the “Reference Image” Feature
6. Don’t Use Branded Words
7. Have Fun!

Don’t Use Artist Names

I know, crazy advice!

Using “in the style of [artist name]” was literally one of my main pieces of advice in my Photorealistic Dall-e Prompting Guide!

Artist keywords are extremely effective in all other AI Image Generators, so it feels counterintuitive to remove them from your Firefly prompt.

However, since Firefly was trained on Adobe’s Stock Library, it doesn’t actually know any artist names.

See how the Firefly AI struggles with Lee Jefferies–an extremely popular photographer keyword in Dall-e–in the Firefly example below.

While Firefly did create a pretty great photo of an old man, it didn’t pull any of the distinct styles from Lee Jefferies’ work.

See how Dall-e replicates the colour, crop, contrast, and composition of Lee Jefferies original work? But Firefly’s image is missing all those distinct characteristics. This shows Firefly has no concept of this artist’s original work.

In my opinion, this is fantastic news!

Firefly gives you the opportunity to generate beautiful photos without having to worry about co-opting YEARS of an artist’s hard work.

Use Adobe Stock for Inspiration

My best piece of advice for prompting with any AI image generator is: understand how the AI was trained.

The more you understand how the AI sees the world, the better you become at speaking its language.

I’ve said it twice already but that’s because it’s important, so I’ll say it again so you can hear me in the back: Firefly was trained off the Adobe Stock library.

This gives you a unique opportunity you don’t get with other AI programs:

You have direct access to the data that Firefly was trained on.

Of course this is a slight oversimplification. We don’t have access to the exact data set used to train Firefly, but Adobe Stock’s search is open to the public.

Anyone could simply navigate to the Adobe Stock website, search for photos similar to the one you’d like to create, and analyze the information provided about that photo.

Try and see what the AI would see! What information is important? What words actually describe the image?

I personally look at the words used in the title, category, or keyword sections and try and use a handful of them in my prompts.

Example

Try combining the keywords from multiple Adobe Stock photos for some really cool results!

Creating really long prompts is where you can really see the AI’s creativity shine!

Prompt: Abstract art collage with a marble bust of a beautiful woman with flowers covering the top half of her face, flows in front, surrealism, contemporary art, poster art, flowers covering the top of her face, lower half of a marble statues face, abstract, pop art, tropical, sculpture of a woman, greek, architecture, historical building, stone, Roman art, retro patterns, layered paper

Unlike other industry leaders like Shutterstock, Adobe Stock encourages its users to upload generative AI images. They even go as far as adding an option in the filter to only show generative images in the results.

Browse Firefly Gallery

Speaking of being inspired by other AI artists’ prompts: Firefly has an interactive gallery of some of the best images generated on the platform with the prompts used to generate them.

You can browse through all the images to see how others created such beautiful pictures!

Take note of the keywords they used, and how the AI pulled those words into the final product. Study the order in which the prompter laid out the information. Are there things in the prompts that aren’t in the final image?

There is a TON you can learn by studying what others have done.

You can take this learning a step further by clicking the “Try Prompt” button that appears when you hover over an image and test the prompt for yourself.

Test what happens when you add new keywords to their prompt. What happens when you change the image’s dimensions or when you add a different style from the sidebar? Better yet, what happens when you remove keywords? How does that change the AI’s understanding?

There is a ton you can learn by pulling apart a prompt that works, and dissecting it until it doesn’t.

You can find the gallery by clicking the “Gallery” tab on the Firefly homepage. Or you can scroll through the gallery that is superimposed on the background of the search page.

Gallery Feed on Firefly Search page

Use Firefly’s Sidebar

I go into the sidebar in detail in my How to Generate Images in Firefly post if you’d like more but for the purpose of this tutorial, we’ll keep it light.

This sidebar solves an issue that plagues all other AI Image generators: it takes out the guesswork of finding keywords that the Firefly AI understands.

You can of course still test out your own words to create more unique images, but using the sidebar opens up prompting to people without an encyclopedic knowledge of art history terms. :p

The sidebar lists keywords for:

• Content-Type: Option to choose between photography, digital art, art, or none.
• Style: Keywords that describe a particular appearance, technique, medium, or movement.
• Color & Tone: Describes the general vibe of the image’s colouring.
• Lighting: Keywords for types of manufactured and natural lighting.
• Composition: Describes the angle and zoom of the lens.

Using a combination of your ideas and the sidebar options allows you to quickly generate anything you can possibly imagine!

Use “Reference Image” Feature

Firefly has this AWESOME unique feature where you can use one of your resulting images as a reference to prompt further images.

So if you get a really good image, but there’s something a little off, you can click the three-dot menu on that image, click “Use as Reference Image”.

Next, this image will pop up with a slider underneath it near your prompt. 

You can click and drag this slider to determine whether you’d like the new images to be more like your prompt, or more like the image.

More Like Reference Image

See how when the slider is all the way to the left, there is only minimal differences between the results and the original image.

Sort of feels like a “spot the difference” game in the newspaper the results are so close.

More Like the Prompt

Alternatively, see the variety in the generated images when the slider is all the way right.

The background, the lighting, the character, and even the color is slightly different, yet you can sense the familiarity to the reference image.

Somewhere in the Middle

I bet you would never guess, but when you have the slider in the middle, the resulting images half follows the image and half follows the prompt.

The Goldilocks Zone

I find having the slider somewhere in between the middle and the prompt to the right creates the coolest variety of images.

It gives you unique options, that still pulls some of the best parts of your reference images!

One of the greatest parts of this feature is that you can change the prompt in between variations of the original image!

I promise you, you will be using this feature A TON!

Don’t Use Branded/Proprietary Keywords

Since Firefly was trained off of open/commercially licensed images, it actually doesn’t recognize any pop culture styles.

For example, check out the difference in how Dall-e handles a prompt for “a cabin in the woods in the style of Harry Potter” compared to Firefly.

Dall-E Example

Prompt: small cabin in the woods in the style of harry potter

The Dall-E image pulls in that unusual architecture, heavy wood-toned, magic vibes that jump off the screen in Harry Potter movies. It even pulls in a toned-down version of the Griffendore red on the door.

Firefly on the other hand generates a photo of a pretty general cabin that I could walk by in my own Canadian back yard.

Firefly clearly doesn’t know what or who Harry Potter is.

Have Fun!

I think the only way I could have ended this even cheesier would have been if I accompanied that title with a cheesy rainbow gif…

Ok seriously though, having fun is important!

Some of the greatest art comes out of a place of play. 

It can be a bit difficult to see the individuality in generative artistry since it isn’t you who necessarily creates the thing, it’s the AI.

But, the AI can’t create anything without you to prompt it! Only you can come up with your specific prompt!

Let your unique self out in this process and HAVE FUN!

Share your work with the hashtag #digitalunicorn for a chance to get featured!