Producing artwork is much more available these days than it has ever been — and that is thanks to AI picture generators. The electrical power to develop spectacular and meaningful artwork just from a single prompt? Signal me up.
But not each text-to-picture generator is produced equally. Midjourney has been my go-to for the final couple of months and I’m a massive fan. Nonetheless, I would nonetheless say that it has some glaring weaknesses, notably in text generation and other nuance.
I’ve been testing it towards other AI picture generators as they all seem to be to preserve enhancing and the up coming one particular on the listing is Adobe Firefly 2. Launched just a couple of days in the past, this new edition claims to have greater nuance and creativity than what the unique edition presented.
Since of its native integration with other Adobe apps, some graphic designers have been providing this new model a go and it is protected to say that the reception has been good. I have attempted it in Photoshop and Adore it. Helps make my lifestyle so considerably simpler.
But how do these resources stack up towards every other?
A Bunch of Distinct Prompts In contrast
To get a complete scope of their creativity, I’ll be dividing the outputs into many classes which includes intent and prompt length. This will aid us get a greater concept on what places they excel at and the place they fail.
Just before I start off, you need to also know that all Midjourney photographs are on the left and Firefly outputs are on the correct.
Practical Photograph
Prompt: The Cliffs of Moher
Midjourney produced a spectacular artwork of the Cliffs of Moher for the duration of a sunset. The artwork is calm and vibrant, but it falls victim to the model’s tendency to overstylize its outputs. If we’re going by pure realism, then Firefly wins. It is not as innovative as Midjourney, but the picture seems like a photograph. You could insert this into Google Pictures when I search the “Cliffs of Moher” and I wouldn’t bat an eye.
Near-Up
Prompt: An intense shut-up of a woman’s encounter for the duration of a celebration
Firefly’s picture has a “stock image” high quality, which helps make sense due to the fact most of the generator’s end users will probably use it due to the fact of the Adobe ecosystem and integration to other apps this kind of as Photoshop, Following Results, and much more.
On the other hand, Midjourney’s output has a great deal much more texture and detail. You can see the imperfections on the woman’s encounter, and her eyes are a great deal much more comprehensive when you zoom in. I only want that Midjourney did not disregard the “party” component in the prompt, but general, this is the clear winner.
Pastiche (Mimicking Other Artists)
Prompt: A quiet evening in the suburbs, in the type of Van Gogh
Adobe Firefly lacks the imagination of a Van Gogh painting, which helps make it search much more like a children’s guide illustration. Midjourney totally understood the assignment. The sky borrows heavily from Van Gogh’s Starry Evening and I like that the street displays the light from the sky. No contest, Midjourney is way greater at imitation.
3D Render
Prompt: A 3D render of a lit candle in a desk
This one’s challenging. I genuinely like each of them. Midjourney leaned in direction of making a practical 3D render, even though Firefly seems much more like a painting. Nonetheless, if I’m to use one particular of these two, I’d most likely choose Midjourney due to the fact of the candle’s information. It also has much more precise lighting than Firefly.
Surrealist
Prompt: A surrealist painting of guys who search like sheep strolling in a occupied Shibuya Crossing
I enjoy Midjourney’s output right here. It is somewhat unnerving, but in a excellent way. There is a great deal of detail in the background, which is essential in establishing the Shibuya Crossing setting. It also aids that the artwork seems critical due to the fact the self-awareness adds a layer of surrealism.
Adobe Firefly has a much more Dadaist get on surrealism. The sheep’s faces search pasted on people’s bodies. There is no cohesion, but it nonetheless has that component of whimsy. Nonetheless, I really don’t like the softness of the background factors due to the fact it helps make it search much more AI than it wants to be.
Abstract Ideas
Prompt: Visualization of language
Asking AI picture generators to develop visualizations of abstract ideas aids us comprehend its nuance. Each Midjourney and Firefly did effectively right here, in my view. The former interpreted language as information inside of a man or woman, whereas the latter understood it as a medium of communication in a neighborhood. This is a clear tie.
Movie Even now
Prompt: A movie nonetheless of a kaiju taking down a constructing for the duration of the evening
Firefly had a weird response to this prompt. Not only does the topic search practically nothing like a kaiju, it also produced a nonetheless of an animation film rather of a dwell-action one particular. There is also no noticeable knocked down buildings. It did not comprehend the context at all.
Midjourney, even so, produced the best movie nonetheless. It seems straight out of a Godzilla film. It is kinetic, action-packed, and practical. This is one particular of my favourite outputs from Midjourney so far.
Architecture
Prompt: A two-storey bungalow in the middle of a quiet discipline
Midjourney has a much more interesting output. The property is possible and someplace I’d in fact enjoy to dwell in. There is also the tiny sneak peeks we get of the house’s interior, which contributes to its realism. Promote this property on Zillow and no one would recognize it is AI.
I’d like you to zoom into Firefly’s property. Observe some thing off about the walls? The property does not make sense. The walls are inconsistent and extremely hard. There is also no window close to the balcony and the property seems so… bland. It is not a excellent architectural style at all, so that is a stage for Midjourney.
Flat Illustration
Prompt: A flat vector illustration of a cute Q-tip
Appropriate off the bat, each these illustrations are quite cute, but neither genuinely seems like a q-tip. Nonetheless, because the leading seems like cotton, I’d choose Midjourney’s artwork for this group, specifically because Firefly’s much more reminiscent of a half-eaten popsicle stick.
Sketches
Prompt: A sketch of a rabbit jumping down a hole
This is one more instance the place Midjourney overstylized its output. The rabbit itself is excellent, but the rest seems chaotic. There is no purpose to include that pop of shade to a sketch and it just contributes to the general mess.
At times, simplicity is crucial — and Firefly demonstrates that flawlessly in this picture. There are no needless additions, leap a easy rabbit jumping out (or rather, into) a hole. It also feels much more “human” than the Midjourney one particular due to the fact it captures freehand artwork greater.
Text Generation
Prompt: The menu of a rustic coffee store
With DALL-E 3 promising greater text generation, I wondered how effectively these two can develop text in photographs. Turns out, not effectively.
From afar, Midjourney is passable. As you zoom in, you recognize how none of the creating in fact seems like phrases. Nonetheless, that is way greater than Firefly, who did not even try to imitate text. Rather, it just went the straightforward route and averted producing a single word.
That explained, I wouldn’t precisely phone any of these winners for text generation. That is why this round is one more tie.
Higher Context
Prompt: A weary guy sporting a blue shirt and black slacks safeguarding his two daughters and one particular son towards falling asteroids in New York City for the duration of a white Christmas evening.
I’ve been making use of Midjourney for months and one particular thing’s for specified: it sucks at substantial-context prompts. Nonetheless, I needed to see how effectively it does towards Firefly.
The reply? Rather effectively. Midjourney managed to adhere to each single detail of my prompt except for the quantity of young children in the picture. It also captured the tone I’m seeking for, which is some thing to think about specifically when I noticed Firefly’s picture.
Firefly not only missed many factors of my prompt, it also missed the mark totally of the prompt’s intent. The household in the photograph seems satisfied and unbothered, and there are no impending meteorites in the sky. This is a clear win for Midjourney.
The Ultimate Verdict
I’d nonetheless use Midjourney above Firefly. It is much more robust at picture generation and, with V6 close to the corner, we can count on Midjourney to get that leap that would push them way over its counterparts.
Total, Midjourney is just much more innovative than Firefly. It is greater at making artwork and, remarkably, greater at comprehending context. That explained, Firefly is nonetheless a excellent different if you are seeking to develop much more practical pictures or you are a graphic designer who’s making use of the Adobe ecosystem.
The best AI image generator is genuinely just the the one particular that fits your type greater. Use this as a manual when you happen to be on the edge of discovering your up coming subscription.