When we're talking about AI image generators, we tend to focus on the end result – the wonderful images we get after just a few seconds. However, we sometimes forget about the prompts that create them. They're the unsung hero behind the scenes: the ones that power creation through the perfect string of words.
Except for the conversational nature of DALL-E 3, most AI image generators follow a specific format: fragmented information, separated by commas or periods, about what you want an image to become. So, if we were to pit two AI image generators in a head-to-head comparison, say Midjourney V6 and Adobe Firefly 2, how would they fare with the same exact prompts?
That's what we're finding out in this article.
Differences Between Midjourney and Adobe Firefly Prompts
Before V6, the key difference between Midjourney and Firefly prompts is that the former was a lot more fragmented and tokenized, whereas Firefly is a bit more conversational. For example, this was a prompt I used for V5 compared to how I'd structure it with Firefly:
video game asset, game map, large lands and lakes, skyrim, cinematic, unreal engine, concept art
Adobe Firefly 2:
Video game asset, created by unreal engine, game map with large lands and lakes, in the style of skyrim, cinematic, concept art aesthetic
Now, with more nuance, I have a lot more flexibility with Midjourney prompts. I can play around with how I structure them since V6 is a lot more forgiving and flexible than V5.
Another difference between these AI image generators is how you define output parameters. Midjourney is a lot more complex since it uses keywords, while Adobe Firefly is more streamlined with a settings block on the right side of its editor. It's just a matter of preference, but the gist of it is this: You can do a lot more with Midjourney, but Firefly gives you better control and is more user-friendly.
I'd go as far as to say that Midjourney and Firefly are actually the two image generators most similar in prompting mechanics if we disregard output parameters. They use the same keywords (for example, "in the style of") and the same delimiters (commas, periods).
Adobe Firefly 2 vs. Midjourney V6: 20 Prompts Compared
There's no better way of showing how Firefly and Midjourney react with prompts than through a comparison. So, here are 20 pairs of images using the same, exact prompts.
Note: All Midjourney V6 outputs are on the left, while the Adobe Firefly 2 images are on the right.
a groom wearing a light gray suit with a sage green tie and pocket square, glamour street medium format photography, portrait, natural lighting
full body shot, a young asian woman with a blonde shoulder-length haircut, loose fitting sweatshirt, translucent layers, soft mist, dreamy atmosphere, beige aesthetics, blue background, analog photography, vogue
2020s fashion, full body of a 23-year-old japanese man sitting, minimalist and comfort fashion, snapshot
a portrait photography of a young woman in the 1990's, unsplash, 90s photography, analog, style of petra collins
A Polaroid photo of a young boy carrying a young Golden Retriever with a date stamp in the bottom white border of "June 16, 1989" in red ink
When Firefly 2 generated the groom picture, I wondered if I was wrong about this AI image generator — that it was actually capable of generating realistic portrait images. Turns out, I'm only half-right. Firefly can create accurate portraits only if it has limited supporting details. Except for the first set of images, Midjourney is the clear-cut better option.
The biggest flaw of Firefly is that it skips over or misunderstands some of the supporting details. Instead of generating a pocket square, it created a flower. Firefly also did not generate an actual Polaroid photo as prompted but tried to simulate the look of a Polaroid photo by placing a square frame around the main subject of the image.
As for the other images, I still prefer V6's interpretations of the prompt. It's more in line with what I have in mind, and it adds a unique flair to the final output.
landscape photography, northern lights in iceland, pastel night sky, evoking a sense of wonder, wide angle shot
view from the top of a mountain during a thunderstorm, with dark and ominous clouds in the background, nikon photography, landscape, trees furiously swaying, overcast, blue lighting
These are different interpretations of the same prompt, and there's something to like about all of them. However, there's something that feels off about the Firefly images. For example, the hills on the first prompt look out of place. I'm also not a fan of having sunlight in the second prompt, as it cheapens the danger of thunderstorms.
Product and Fashion Mockups
commercial photography of a windbreaker with pastel 80s vaporwave aesthetic, woman, streetwear, fashion, mamiya rb67
product photography, craft beer, beautiful lighting, vintage aesthetics, earthcore, ultra-detailed, atmospheric
These are both good entries, but I still prefer Midjourney's output. If we're talking about adherence to the prompt, they both performed well, but Midjourney's improved creativity just puts it above and beyond Adobe Firefly for now.
a man in a suit looking at an ominous giant green moon, dark background, inspired by pink floyd, album art aesthetics, linocut print
an illustration of a girl in awe, inspired by stop-motion animation, contemporary fairy tale, magical realism, gelatinous forms, simple minimalism
a lazy sunday afternoon, andrew tomine, pastel colors, defined outlines, comfort and warmth, apartment aesthetics
art by Studio Ghibli, a surreal scene of a beach with bioluminescent jellyfish in the sky, night, psychedelia, minimalism, fluorescence, vibrant colors
a man with blue eyes looking at the sky, view from above his face, risograph, anime painting
a manga sketch of a young woman with a melancholic expression, chiaroscuro, pale yellow background, defined outlines
The only output here that's relatively close in quality are the jellyfish artworks. Besides that, Firefly's images are too one note. Even in artwork, they still have that look of stock images, complete with big round eyes and a style that resembles children's books. The one that didn't have a person as its subject (lazy sunday afternoon) completely disregarded the "apartment" part of the prompt and wasn't generated logically.
witch crafting game logo, witch, magical spells, dagger, simple, minimalistic, round, in the style of Paul Rand
a simple vector logo of a white tulip, inspired by van gogh paintings, minimalist
a clean logo, avocado, postmodern, massimo vignelli, vivid colors
It's one thing to disregard a supporting detail, but to completely miss the main point of a prompt doesn't bode well for Midjourney. In these three sets of images, only the first one from Firefly passes as a logo. The other ones look more like a complete artwork, especially the tulip.
High Context Prompts (2)
A wise old wizard conjuring a glowing spell with ancient runes swirling around his hands. He has long grey hair and beard and wears purple robes covered in arcane symbols. Shadows dance around him as magic flows from his fingertips. His wrinkled face is etched with concentration while casts the mystical incantation. He's on top of a tower. In the style of vintage dark fantasy illustration and lithography.
minimalism and surreal illustration, Nick Veasey, extremely close-up shot, landscape of the cliffs of moher, light tracing, glare lighting, linocut print, minimalism, line art, curve linked smoothly, dry brushing style‚ impasto, fluorescent color, colorful tones, ripples of color diffusing, marker coloring, printmaking style, layered curve, fujicolor superia X-TRA
When I created these images, I purposefully created two types of high-context prompts: one that's conversational and one that's tokenized with lots of variables to consider. Both these images performed well with the former, although Firefly wasn't able to create it in the style of dark fantasy or lithography illustration.
On the other hand, the difference in the tokenized long prompt is more stark. Firefly ignored more than half of the supporting context.
If we're talking only about creativity, Midjourney V6 is miles ahead of Firefly 2. There are many reasons that contribute to this, including the main goal of the company behind them, with Adobe spreading their manpower over multiple products and Midjourney only having their AI image generator as their sole focus.
However, there's still a lot to like with Firefly 2. For one, it's completely free. It also supports direct integration with other Adobe products, which makes inpainting possible with Photoshop, InDesign, and more.
As for nuance and prompt understanding, Firefly and Midjourney are surprisingly close with shorter prompts, but the former can't handle high-context ones without missing some key information.
So, which would I recommend? If you have the money, go for Midjourney. That said, Firefly 2 is a damn good free alternative.
If you're interested in knowing more about Midjourney and Firefly, I highly recommend reading our other comparison articles. Have fun!