Firefly vs Midjourney vs DreamStudio: The Battle Royale of AI Image Models

Gone are the days when we thought art could only be created by artists. Or that images could only be captured by photographers. In this new age of generative AI, numerous AI image generators on the horizon can pick up your thoughts and create something that's very close to your imagination. But in the plethora of options available, which one should you pick? And which one suits your needs the best?

‍

We did a deep dive comparing three most popular AI image generators available right now using three different prompt levels — Adobe's Firefly (Beta), Midjourney, and DreamStudio (depth-guided stable diffusion model, finetuned from SD 2.0-base).

‍

An L1 prompt means the simplest prompt, using few words, focusing only on the main components of the image.

‍

L2 prompt means we ask the AI tool to add more details to the basic L1 prompt.

‍

L3 prompt means we add more details to the prompt, adding intricacies and secondary image components that will enhance the overall look.

‍

Let's dive in!

‍

1. Landscape

‍

L1 Prompt: A place in Wales, it’s day break, there’s a brook flowing nearby

‍

‍

DreamStudio's output is the closest to a real life image yet seems very bland. The other two tools do a much better job at creating images that have more details and nuances in them even with such a basic prompt.

‍

L2 Prompt: A place in Wales, it’s day break, there’s a brook flowing nearby, intricate, high definition, soft lighting

‍

‍

Again, DreamStudio seems to be working from a very narrow point of view. Whereas, Firefly's output seems like another version of the L1 prompt instead of a high definition, enhanced version.

‍

L3 Prompt: A place in Wales, it’s summer’s last day break, there’s a brook flowing nearby, a tucked away corner that few tread upon, gloomy yet cozy weather intricate, high definition, soft lighting

‍

‍

Midjourney seems to do interpret the added secondary details much better. It's the only output that reflects 'tucked away corner' and 'gloomy yet cozy'.

‍

Verdict: Midjourney > Firefly > Dreamstudio

‍

2. Lyrics interpretation

‍

L1 Prompt: Light pink sky up on the roof, Sun sinks down, no curfew

‍

‍

We wanted to see how well each tool could interpret an artist's lyrics and generate images that, till now, we could only conjure up in our minds. We chose Taylor Swift's song It's Nice To Have A Friend and picked these lyrics:

‍

Light pink sky up on the roof

Sun sinks down, no curfew

‍

This basic prompt was interpreted by each tool as we'd expected. DreamStudio's output seems the closest to reality yet Firefly and Midjourney do a good job at highlighting the roof and Sun sinking down in the sky. Though Firefly's Sun seems a bit distorted.

‍

L2 Prompt: Light pink sky up on the roof, Sun sinks down, no curfew, high definition, ultra realistic, intricate

‍

‍

As we leveled up, DreamStudio's output was pretty standard. Instead of more definition, the output just looks like a different version of the L1 prompt. All three tools, however, have interpreted the prompts at a surface level and we still don't see any essence of the song incorporated in any of the six outputs yet.

‍

L3 Prompt: Light pink sky up on the roof, Sun sinks down, no curfew, Taylor Swift song, Lover album vibes, high definition, ultra realistic, intricate

‍

‍

It's only when we mention the artist's name that the outputs incorporate human figures. However, DreamStudio fails to do that again.

‍

Verdict: Midjourney > Firefly > DreamStudio

‍

3. Anime

‍

L1 Prompt: Fullmetal Alchemist Brotherhood anime style, Roy Mustang using his fire power

‍

‍

While Firefly did show Roy Mustang using his fire power, the composition of the image is quite poor. See how both hands aren't fully formed. Plus, Roy Mustang doesn't look anywhere close to how he looked in the anime. DreamStudio does any even shoddy job at this. The composition — especially the face — is terrible and he's not even shown using any fire power.

‍

Midjourney, however, aces the prompt.

‍

L2 Prompt: Fullmetal Alchemist Brotherhood anime style, Roy Mustang using his fire power, high definition, ultra realistic, intricate

‍

‍

While Firefly and DreamStudio both seem to be showing a person and fire in anime style, the composition is still super unsatisfying. Firefly did not compose the face well and seems to be adding a devilish touch to Roy Mustang's persona.

‍

Midjourney's output is very intricate. We can also see improvements in color grading and addition of small details on his uniform.

‍

L3 Prompt: Fullmetal Alchemist Brotherhood anime style, Roy Mustang using his fire power, fiery explosions, thundering clouds, intense and gloomy facial expression, high definition, ultra realistic, intricate

‍

‍

For some reason, Midjourney turned the image super gloomy and dark. Instead of showcasing thundering clouds and an 'intense and gloomy expression' on his face, it seems to have incorporated it throughout the image.

‍

Verdict: Midjourney > Firefly > DreamStudio

‍

4. Fantasy

‍

L1 Prompt: Female warrior, archer, controls elements of the Earth

‍

‍

Midjourney does agood job. Firefly and DreamStudio's outputs seems quite similar. However, DreamStudio does a better job at showing the bow and arrows of the archer.

‍

L2 Prompt: Female warrior, archer, controls elements of the Earth, cinematic lighting, intricate background, photorealistic, high definition

‍

‍

L3 Prompt: Female warrior, archer, controls elements of the Earth, the Moon and Sun behind her, she’s aiming at someone, fiery eyes, intense expressions, female rage, cinematic lighting, intricate background, photorealistic, high definition

‍

‍

Firefly has improved leaps and bounds in showcasing the archer — with its bows and arrows — but there seems to be a continuity issue where the archer is pulling the arrow back. Even though DreamStudio's render of the moon is very basic, the archer pose seems better.

‍

Verdict: Midjourney > Firefly = DreamStudio

‍

5. Animals

‍

L1 Prompt: Cows grazing on the Dolomites hiking trail, soft summer sun

‍

‍

Firefly and DreamStudio's outputs are quite artistic while Midjourney's seem more realistic. Interestingly, DreamStudio does a great job at interpreting 'Dolomites' correctly – rolling grassy hills with jagged, saw-edged ridges and rocky pinnacles. See how the grass in Midjourney's output is super dry and patchy.

‍

L2 Prompt: Cows grazing on the Dolomites hiking trail, soft summer sun, cinematic lighting, ultra detailed, photorealistic

‍

‍

L3 Prompt: Cows grazing on the Dolomites hiking trail, they have bells around their necks, soft summer sun, children playing nearby, tiny flowers sprouting from the ground, cinematic lighting, ultra detailed, photorealistic

‍

‍

Only Firefly added both children and flowers as secondary details. And Midjourney's outputs definitely have higher quality at all three instances. However, DreamStudio still does a better job at staying true to the core of the prompt i.e. cows in Dolomites. So we're going to give more importance to how well each tool interpreted each part of the prompt instead of just image quality.

‍

Verdit: DreamStudio > Firefly > Midjourney

‍

6. Flora

‍

L1 Prompt: Serene Zen garden, exotic plants

‍

‍

The outputs from all three tools is pretty standard.

‍

L2 Prompt: Serene Zen garden, exotic plants, very intricate, hyper realistic, high definition

‍

‍

We like Firefly's output the best as of now since it build a 'garden' scene much better – the stony path, smooth pebbled rocks, exotic plants, serene weather.

‍

L3 Prompt: Serene Zen garden, exotic plants, glowing crystalline structures, a waterfall cascading into a bioluminescent pool, very intricate, hyper realistic, high definition

‍

‍

DreamStudio falls quite flat at all three levels while both Firefly and Midjourney add the secondary details quite well.

‍

Verdit: Midjourney > Firefly> DreamStudio

‍

7. Van Gogh

‍

L1 Prompt: Huge ballroom style room with Salsa dancers, inspired by Van Gogh

‍

‍

DreamStudio was the only one that could incorporate a direct connection to Van Gogh's painting style in the ballroom floor. However, it completely overlooked the addition of any Salsa dancers to the output.

‍

L2 Prompt: Huge ballroom with Salsa dancers, inspired by Van Gogh, bold colors, expressive, post-impressionist, intricate, high definition

‍

‍

But it completely butchered the prompt as we added more details.

‍

L3 Prompt: Huge ballroom with Salsa dancers, inspired by Van Gogh, night light ushering through large windows, mirrorballs, glitter, shimmer, high energy, bold colors, expressive, post-impressionist, intricate, high definition

‍

‍

While DreamStudio continued to use blue and yellow colors – synonymous to Van Gogh – the picture quality continued to deteriorate. The other tools, however, continued to slowly incorporate touches of Van Gogh's painting style while also keeping picture quality high.

‍

Verdit: Midjourney > Firefly > DreamStudio

‍

8. Architecture

‍

L1 Prompt: A buzzing and lively underwater metropolis, vintage architecture

‍

‍

Out of all three, Firefly's interpretation of a metropolis is on point. Midjourney's output certainly incorporates victorian touches but it's far from a metropolis. DreamStudio on the other hand is highly dissapointing.

‍

L2 Prompt: A buzzing and lively underwater metropolis, vintage architecture, high definition, hyper realistic, intricate, cinematic lighting, 35mm

‍

‍

Again, Firefly's output is the closest to a matropolis while Midjourney stays true to the victorian architecture.

‍

L3 Prompt: A buzzing and lively underwater metropolis, vintage architecture, marine life swims freely in the city’s architecture, citizens commute using personal submersible vehicles, high definition, hyper realistic, intricate, cinematic lighting, 35mm

‍

‍

Verdit: Firefly > Midjourney > DreamStudio

‍

9. Pop culture

‍

L1 Prompt: If Harry Potter was a Marvel superhero

‍

‍

This is perhaps the most hilarious one. Harry Potter is a cis-male and possibly the most iconic and popular figure in the entertainment industry. Not to Firefly it seems! It completely failed at recognizing Harry Potter and gave us a random output.

‍

Midjourney, again, does a fantastic job at image composition. However, Harry Potter's eyes seem off and the facial age of this person seems much older than what you'd expect.

‍

L2 Prompt: If Harry Potter was a Marvel superhero, high definition, hyper realistic, intricate

‍

‍

While DreamStudio's image composition isn't the best, it's the one that gets closest to Daniel Radcliffe's face.

‍

L3 Prompt: If Harry Potter was a Marvel superhero, wand ready, midnight moon, eerie, high definition, hyper realistic, intricate

‍

‍

Verdict: Midjourney > DreamStudio > Firefly

‍

Our Final Choice

‍

Clearly, how well an AI text-to-image tool works depends heavily on the prompts it is fed. But if we consider only the categories here and the outputs each tool gave us, here's what we think.

Verdict: Midjourney > Firefly > DreamStudio