In the hullabaloo caused by the flurry of announcements at the recent Google I/O conference, some stories were overlooked that, on any other day, would have been front page news. One that surely fits that category is Adobe’s partnership with Google, bringing the Adobe Firefly text-to-image generator AI into Google Bard, and by extension into all of Google’s products.
Firefly is currently in beta, with waiting list times for admission to the service rumored to be as long as a month, but when the service goes public it will have an enormous impact on Google’s billions of users. What’s more, although Firefly seems quite basic at first glance - especially when compared to rivals such as Midjourney - it packs one hugely useful and practical feature that makes it an essential tool for professionals looking to create specific images.
A FORTUNATE ASPECT
One of the key ways Firefly differentiates itself from its rivals is by offering a wide variety of optional tags and modifiers you can add to your prompt. You can perform similar tricks with other text-to-image generators, but services like Midjourney and Image Creator require you to enter the instructions as part of the text prompt; it’s fiddly, time consuming, and depends on you knowing what options are available to begin with. In Firefly, you pick your modifiers from menus, and it’s a lot less hassle.
This story is part of our weekly briefing. Sign up to receive the FREE briefing to your inbox.
In the hullabaloo caused by the flurry of announcements at the recent Google I/O conference, some stories were overlooked that, on any other day, would have been front page news. One that surely fits that category is Adobe’s partnership with Google, bringing the Adobe Firefly text-to-image generator AI into Google Bard, and by extension into all of Google’s products.
Firefly is currently in beta, with waiting list times for admission to the service rumored to be as long as a month, but when the service goes public it will have an enormous impact on Google’s billions of users. What’s more, although Firefly seems quite basic at first glance - especially when compared to rivals such as Midjourney - it packs one hugely useful and practical feature that makes it an essential tool for professionals looking to create specific images.
A FORTUNATE ASPECT
One of the key ways Firefly differentiates itself from its rivals is by offering a wide variety of optional tags and modifiers you can add to your prompt. You can perform similar tricks with other text-to-image generators, but services like Midjourney and Image Creator require you to enter the instructions as part of the text prompt; it’s fiddly, time consuming, and depends on you knowing what options are available to begin with. In Firefly, you pick your modifiers from menus, and it’s a lot less hassle.
It’s still extremely powerful, though! A great illustration of why comes with what seems like the most basic choice of all: aspect ratio. Firefly currently offers four options, (square, landscape, portrait and widescreen), and your choice will drastically affect the final outcome of the image, as it defines the amount of blank space the AI has to account for.
Some fireflies with the settings menu on the right. These fireflies are "Art", "Vibrant Color", "Golden Hour", "Blurry Background", and "Chaotic"!
There are also dropdown menus for you to specify “Content Type”, “Color and Tone”, “Lighting” and “Composition”. Some of these are more effective than others, with many of the composition settings in particular being ignored, but the Content Type settings are extremely useful. You can specify Photo, Graphic or Art settings, and this will influence all subsequent image generation. If you’re looking to make use of the more creative settings, we recommend setting Content Type to Art first.
STYLING IT OUT
That’s five of the six options in the menu bar covered, but the sixth is where the fun happens. Under “Styles”, you will find 63 (at the time of writing) aesthetic toggles that you can apply to your prompt, giving you a great deal of control over the final output. These are effective, quick to use, and can easily turn mundane prompts into exciting images, but take care to avoid clashing toggles! Make sure your selections make sense together. Firefly will still produce images if you combine “layered paper” with “wood carving”, but they are not likely to be what you hoped for.
Ordinary pointillist ducks (left) versus bioluminescent pointillist ducks (right).
If you find the 63 provided styles limiting, don’t forget that you have a text field where you can directly prompt the AI with requests. It’s useful to bear in mind here that Firefly has been exclusively trained on images that are either owned by Adobe, or are out of copyright, and while it’s impossible to be sure what Adobe’s massive image library brings to the party, we learn a lot more from the second fact. It means that you can prompt Firefly to borrow influence from the style of any visual artist active until roughly 1920, simply by adding “in the style of [artist name]” to your prompt. You can also attempt this with more contemporary artists, but the probability of getting a recognisable result is greatly diminished, and where possible you should prompt for art movements and styles rather than the work of individual artists. One other note here is that we’ve found adding appropriate style toggles can improve the efficacy of text prompts; for example, you’ll get nice results if you ask for “a family of ducks in a pointillist style”, but if you add the bioluminescence toggle, those ducks are really going to pop!
REFERENCE: THE NEW HOTNESS
Tucked away in the corners of your generated images are more menu options, including two that are vital to prompt refinement. You’ll find the first of these if you hover over the top left corner of an image: “Show Similar” produces new variations on the selected image, a function familiar to any who have used similar services before.
The second option is much more powerful, and where Firefly distinguishes itself at the cutting edge of text-to-image AI tech. In the menu found at the top right of each image - because Adobe believes you can never have too many invisible menus! - there’s an option to select the output as a “reference image”. This will then place a thumbnail of the reference image on the left of the prompt field, and by clicking on it you can access a slider to set the emphasis from “reference image” to “prompt”. By altering the prompt and adjusting the level of emphasis Firefly places on it, relevant to the reference image, you can achieve a substantial degree of control over subsequent iterations of the output. It’s an incredibly useful feature that helps nudge the generative AI process away from dice rolling and towards more structured creative workflows.
Reference images in action!
Invisible menu spam aside, Firefly has introduced other welcome quality of life improvements that rivals could learn form. One welcome feature is the ease of navigating through your work; because Firefly is (for the time being) browser-based, your browser’s forward and back buttons will take you through your prompt history, allowing you to quickly scan your completed work.
TEXT TO TEXT
A quick word on Firefly’s other AI apps: there are three, and in addition to the text-to-image service you will find “Text effects” and “Recolor vectors”. The latter is a bafflingly underwhelming tool that seems to have escaped from Adobe Illustrator, we shall not discuss it here. Text effects, however, attempts to solve a longstanding problem with generative images, by allowing you to create display text and apply extravagant visual effects to the AI-generated lettering. It’s a useful tool that is currently only let down by its limited number of fonts, all of which are ugly. In time we expect it will become very useful, but for now we have just one Text Effect tip, which is that you can also use it to recolor emojis!
HACKER’S CORNER
Finally, what guide to text-to-image generators would be complete without some speculation about secretive means of gaming the prompting algorithm? Like all LLMs, Firefly’s prompts can be enhanced with “flags” that alter its outputs in specified ways. We only know of two confirmed flags (or as Adobe prefers to call them, variables); in Firefly’s text-to-image prompts Adobe staff have demonstrated the use of [avoid=background], and in the text effects app they have revealed the [outline-strength=x] variable.
Adobe staff have described these variables as “buggy”, and for this reason, the company has not released a complete list of other options that can be tweaked. Nonetheless, knowledge of the [variables format] gives us the opportunity to search for other variables, and a better way to frame our prompts. Bear in mind that even if your prompt refers to variables that don’t exist, the prompt interpreter is still going to react to it, it’s merely going to ignore the [square brackets] and treat it like any other text.
This story is part of our weekly briefing. Sign up to receive the FREE briefing to your inbox.
Gadgets & Gifts
Put your spy skills to work with these fabulous choices from secret notepads & invisible inks to Hacker hoodies & high-tech handbags. We also have an exceptional range of rare spy books, including many signed first editions.
We all have valuable spy skills - your mission is to discover yours. See if you have what it takes to be a secret agent, with our authentic spy skills evaluation* developed by a former Head of Training at British Intelligence. It's FREE so share & compare with friends now!