Gold Penguin Logo with Text

Adobe Firefly 2, Midjourney, and DALL-E 3 Prompt Accuracy & Quality Compared

With the release of Adobe Firefly 2, there's a third viable competitor to the already existing DALL-E and Midjourney image generators. I tested a few prompts with each and put them side by side to see how well the prompt was interpreted. Here's what I found
Updated October 23, 2023

I'm not exaggerating when I say that I use Midjourney every day. Scroll up. Do you see that featured image there? That's from Midjourney.

When you've been using something long enough, it's easy to notice the quirks. The little things you wouldn't have noticed initially become more apparent especially when you compare things side by side.

A few days ago I wrote up a comparison between DALL-E 3 and Midjourney but had to use sample screenshots as my source. But since this is the world of AI, things move fast.

Today I was able to test DALL-E 3 through Bing Create to use my own examples. On top of that, Adobe just released a new version of their image generation model called Firefly 2.

So how do these 3 actually compare when looking at the little things. Is there one you should use if you're looking for a certain vibe over another?

What are DALL-E 3, Midjourney, and Adobe Firefly?

Launched in October 2023, DALL-E 3 is OpenAI’s newest AI image generation model. This evolution of DALL-E 2 has a better prompt understanding than its previous versions and works so much better with text.

Midjourney is known for its detailed outputs, extensive customization through prompt engineering parameters, and nuance. It operates out of annoying discord, but is generally regarded as the best AI image generation tool.

Adobe Firefly 2 is the next generation of Adobe's generative AI model for imaging that came out this week. This model can generate better-quality images super easy like other tools.

Adobe Firefly 2 Landing Page

Adobe Firefly vs. Midjourney vs. DALL-E 3: Output Comparison

Photorealistic Portraits

Prompt: a close-up portrait of a tired college student

A close-up portrait of a tired college student

If I were to pick a winner, it's definitely Adobe Firefly. It's realistic, the facial expression is clear but subtle, and it has excellent lighting. On the other hand, Midjourney's output is also pretty good, but it's softer — almost like a painting. There also seems to be a rendering issue in the book, but it's practically invisible unless it's pointed out.

DALL-E's generated image is… well, it definitely understood the prompt. It overemphasized the student's exhaustion. Those eyebags are too unnatural to be realistic, plus it has a "stock image" quality to it. 

It's also worth mentioning that none of these images have an uncanny valley feeling, which is a huge plus in my book.

Design & Architecture

Prompt: a wide angle view of a stylish brick wall manhattan-style loft with a sunken living room

A wide angle view of a stylish brick wall Manhattan-style loft with a sunken living room

There's a similarity between these three images: none understood the intent of my prompt completely. They all created a Manhattan-style loft but struggled in making the sunken living room part.

Adobe Firefly has the best lighting of them all. I particularly love how well the shadows correspond with where the light comes from and how well they blend together. It looks the most realistic for a New York apartment.

Meanwhile, Midjourney's biggest asset is its attention to detail. From the books to the paintings on the second floor, it's everything you expect from a typical loft-type apartment.

Again, I'm disappointed with DALL-E. It went overboard with the lighting and had a softer texture. It also has too many living rooms — which is funny & ironic because it's the only one with a sunken section, but it's not one of its four rooms.

Landscape

Prompt: wildflower meadow sunset landscape

A wildflower meadow sunset landscape

Adobe Firefly's output is lifelike, but it's too similar to stock images of meadows online. You could've also missed this, but try zooming in. Do you notice that none of the flowers is rendered properly? It's not noticeable from afar, but the wildflowers appear to be glitching out.

On the other hand, the colors really pop in Midjourney's meadow. However, Midjourney tends to stylize its outputs more; you can clearly see that here. If you want a painting rather than a realistic photo, go for Midjourney.

DALL-E's image feels majestic like it's a still from a fantasy movie. It's warmer, and while it's not the most colorful, it has a detailed texture. It's perfect, which makes it my favorite so far.

Surrealist

Prompt: a surrealist painting of a large firefly in a house made of denim

A surrealist painting of a large firefly in a house made of denim

Adobe Firefly's output heavily borrows from children's books. It showed a great understanding of the prompt, but it's not quite what I'm looking for.

Midjourney gave me exactly what I wanted: that whimsy feeling of surrealist art. It perfectly combines real-world images with fantastical concepts. I would've loved it better if it's set outside, but I like that even the firefly is made of denim.

Another small detail I appreciate is adding other elements that make it weirder, including the hands on the firefly's legs and the boarded-up windows.

DALL-E has a different take, but I appreciate it as much as Midjourney's. It created a narrative by making the denim house pants and the firefly's trying to fit into it.

There's also the little things like the two moons and the pocket windows. It feels more coherent as an art, which makes this a win for DALL-E 3.

I think the issue here was my prompt wasn't super clear on what exact parts of the picture should be what. That's why each of these had its own vision of what it thought I wanted.

Abstract Concepts

Prompt: the visualization of infinity

Visualization of infinity

Infinity is impossible to create, but these three were wildly different but great attempts. Adobe Firefly had a more algorithmic approach to infinity. It's similar to visualizations of the Fibonacci sequence, but this one is more expansive and has more character.

The Midjourney one has a lot of storytelling potential, making it better as a storyboard than standalone art. Meanwhile, DALL-E is more psychedelic, like you're getting swallowed by a vortex of colors. It looks like an intricate tie-dye shirt.

Pastiche (Art that imitate other art)

Prompt: women fighting for equality in the style of dadaism

Women fighting for equality in the style of dadaism

Adobe Firefly has a sound output, but it unfortunately missed the mark. It doesn't look like any Dada art. I also tried tweaking the prompt a bit to get a more accurate result, but the results I get are always similar to this one.

Midjourney and DALL-E succeeded where Firefly failed: they understood the context. Their artwork emulates Dadaism to a tee.

Midjourney leans toward collage art, similar to the famous Hannah Höch. On the other hand, Marcel Duchamp seems to be the main inspiration for DALL-E's art. Both artists were prominent during the Dada Movement.

Flat Design

Prompt: a flat vector illustration of an ai assistant

A flat vector illustration of an ai assistant

Adobe Firefly, once again, failed to understand the prompt. The output itself is fine; it's still vector art — but where's the AI assistant? It's a shame since I actually prefer the look of Firefly's artwork over the other two.

Midjourney and DALL-E's output has a more conventional vector art look. The former's illustration is better for presentations and pitch decks, while the latter is perfect for websites.

In Summary...

AI IMAGE GENERATOR

ADOBE FIREFLY 2.0

MIDJOURNEY V5.2

DALL-E 2

Usability & Accessibility

You have to enter a prompt first before you can access customization tools for aspect ratio, color, tone, lighting, and more. It can also be integrated with other Adobe products such as Photoshop, After Effects, etc.

Midjourney doesn't exist as a standalone software but you can use it through the Midjourney bot in Discord. Parameters can be tweaked before you submit a prompt, but you must study their extensive documentation first. It has a steeper learning curve than the other two.

You can use it on the DALL-E website, through ChatGPT Plus/Enterprise, or with Bing Image Creator. Generated images can be tweaked after the prompt.

Output

When you need stock image for your content, Adobe Firefly is your best bet. Among the three, it produces the most lifelike images.

Midjourney prefers stylized outputs over realism. Almost every image generated by Midjourney looks like a piece of art or an animated movie still. While you can use this tool for your blogs, its main demographic should skew towards enthusiasts.

You should expect simple but effective images from DALL-E. They're not too loud nor muted in color. However, it tends to overcompensate on some prompts, leading to some cartoon-ish results.

Text Generation

Nothing impressive, but it's definitely capable.

Non-existent. Unless you're rendering single letters, you can't expect Midjourney to generate accurate text for images.

The best among the three.

Turnaround Time

The fastest AI image generator I've tried so far.

Almost 1.5x slower compared to the other two.

Just shy of a minute average per generation.

Context Understanding & Nuance

Shows some mistakes, even with low-context prompts.

Generally fine with low-context prompts but struggles with high-context.

Exceptional nuance and context understanding. It rarely misses a line in the prompt.

Bugs

I haven't had any yet.

Midjourney can sometimes get stuck on generating an image, which could create a backlog.

I haven't had any yet.

OVERALL

Best for realism.

Best for creating artwork.

Best for high-context prompts.

In a nutshell, I found Adobe Firefly to be the best for realistic image generations, especially ones you need for photo editing or stock images. Since Midjourney tends to stylize its generated images, it's better for artwork or imitating one. If you're more concerned with generating an image with precision when it comes to prompts, give DALL-E 3 a try.

So Now There's 3 Actual Competitors

I’m glad that other AI image generators are becoming more nuanced and creative. Competition is always healthy. Not only does it give users more options, but it also forces a product to evolve. I think we'll look back in a year and laugh at how "bad" these generations actually were.

Adobe Firefly 2 and DALL-E 3 are both worthy opponents for Midjourney because they each excel in specific aspects.

Firefly is best at creating realistic images, Midjourney excels at complex artwork, and nothing could beat DALL-E 3’s understanding of prompts (and text generation).

Want To Learn Even More?
If you enjoyed this article, subscribe to our free monthly newsletter
where we share tips & tricks on how to use tech & AI to grow and optimize your business, career, and life.
Written by John Angelo Yap
Hi, I'm Angelo. I'm currently an undergraduate student studying Software Engineering. Now, you might be wondering, what is a computer science student doing writing for Gold Penguin? I took up studying computer science because it was practical and because I was good at it. But, if I had the chance, I'd be writing for a career. Building worlds and adjectivizing nouns for no other reason other than they sound good. And that's why I'm here.
Subscribe
Notify of
guest

0 Comments
Inline Feedbacks
View all comments
Join Our Newsletter!
If you enjoyed this article, subscribe to our newsletter where we share tips & tricks on how to make use of some incredible AI tools that you can use to grow and optimize a business
magnifiercross