MEGACOMPARISON February.

Four months later, we’re back to review the most popular generative image AIs.
In this time, we’ve only seen one major update: Midjourney’s V.6 model.

The methodology used is as follows:

  1. Prompts are created using text-to-text generation (ChatGPT) without supervision.
  2. Images are not reviewed or corrected.
  3. No additional parameters are provided to the AI beyond the image format.
  4. We will score and provide feedback for each generated image.
  5. A final score and verdict will be given.

Let’s dive in!

/1/ HYPERREALISTIC PHOTOGRAPHY

A hyperrealistic portrait of a young woman in her twenties, with smooth skin, blue eyes, and straight black hair.
prompt: a hyperrealistic portrait of a young woman in her late 20s with smooth skin, blue eyes, and straight black hair.
Midjourney 6 🟢
DALL-E 3 🔴
Adobe Firefly 2 🟠

Midjourney continues to depict people as “too perfect,” which detracts from realism. Even so, the V6 version has shown significant improvements in composition and detail. Although this image feels a bit over-polished, MJ would undoubtedly be my top choice for hyperrealistic photography.

DALL-E 3 still struggles to produce high-quality realistic images. It looks like a doll.

Firefly has a massive Adobe Stock database, which makes its images almost indistinguishable from reality. Even so, I rank it behind Midjourney due to its lack of versatility. Creating a realistic portrait is very straightforward, but if you’re looking for something more specific, it will be harder to achieve.

/2/ MOCKUP

Illustration of an outdoor billboard advertisement located on a busy street with 'CREATIVE DIRECTION'.
prompt: Illustration of an outdoor advertising banner placed on a busy street with 'DIRECCIÓN CREATIVA'
Midjourney 6 🟠
DALL-E 3 🟢
Adobe Firefly 2 🔴

Everything surrounding the billboard is well-crafted, and once again, Midjourney demonstrates its great ability to create photographic images.

On the other hand, what appears inside the billboard is unintelligible.

Although this image looks a bit more 3D, DALL-E usually delivers good results when creating mockups, although sometimes with low precision in the text—nothing that can’t be fixed with a bit of editing.

Four months later, Adobe Firefly has made no progress in this field.

 

/3/ GASTRONOMIC PHOTOGRAPHY

Foto hiperrealista de tres hamburguesas distintas sobre una superficie oscura reflectante, con un muro de ladrillos marrones rústicos en el fondo.
Hyperrealistic photo of three distinct burgers on a reflective dark surface, with a rustic brown brick wall in the background.
Midjourney 6 🟢
DALL-E 3 🔴
Adobe Firefly 2 🟠

A very realistic image; the burgers look real, although the table reflection seems overly exaggerated (this could be corrected). The depth of field and background are well-executed. Good result.

Very 3D, unnatural, with oddly shaped food. DALL-E fails at almost everything related to photography.

It doesn’t look very realistic. Although the result is better than DALL-E, it still isn’t a marketable product. The ingredients and lettuce look like plastic.

/4/ PRODUCT PHOTOGRAPHY

Photograph of the perfume 'Cítricos de Sevilla,' a vibrant and refreshing fragrance.
prompt: Photograph of 'Citrus of Seville' cologne, a vibrant and refreshing fragrance. The product is showcased with a backdrop of lush Andalusian gardens, and the foreground features blood oranges to emphasize the citrus notes.
Midjourney 6 🟠
DALL-E 3 🟢
Adobe Firefly 2 🔴

Midjourney once again stands out in capturing reality photographically. However, the perfume bottle is a bit strange, as is the red orange. The composition doesn’t place the product at the center, and the text is not legible.

Although it doesn’t look very realistic, both the composition and the text are well-executed. It’s not a perfect result, but it could be improved. The composition is very good; I would remove the seated figures.

A fairly realistic image with significant errors. The orange color inside the bottle and the lack of a label and text are the most glaring issues.

/5/ ARTISTIC ILLUSTRATION

An illustrated poster in the 1920s style, inspired by Peaky Blinders, featuring a 3D cat as the main character.
prompt: An illustrated poster in the style of the 1920s, inspired by Peaky Blinders, featuring a 3D cat dressed in a tailored suit and a flat cap, embodying a cunning gang leader. Positioned against a dark and foggy backdrop of post-war Birmingham, the cat holds a gold coin in one paw, projecting a look of authority and mystery.
Midjourney 6 🟠
DALL-E 3 🟢
Adobe Firefly 2 🔴

I don’t love it, but it passes. The background is very well done, but the cat’s expression isn’t great. It doesn’t quite look like an illustration, even though that’s what was requested.

Very doll-like but visually appealing. It shows the Tower of London when the prompt specifies it should be Birmingham. DALL-E remains the benchmark for creating illustration and 3D.

Poor result, it looks like something from an earlier generation of GenAI.

 

/6/ TEXT

Create a text image with 'OYSTERS', 3D, metallic violet-purple.
prompt: create text image "OYSTERS" impacted a logo, use 3d, violet-purple metalic
Midjourney 6 🟠
DALL-E 3 🟢
Adobe Firefly 2 🔴

Although it has made significant progress in generation, there is little control over what it produces, and it makes mistakes like omitting letters. I didn’t ask for oysters to appear, but it included them because it deemed it appropriate.

Very good result. A solid high grade for DALL-E.

Adobe has a specific tool for generating text, but it limits you to specific fonts and formats. It essentially fills in pre-existing text, and the results are not good. If you ask Firefly, as you can see, it doesn’t even respond properly.

/7/ ARCHITECTURE

house, modern, panoramic windows, black concre
prompt: house, modern, panoramic windows, black concrete
Midjourney 6 🟢
DALL-E 3 🟠
Adobe Firefly 2 🔴

Quite a good result, although the texture of the concrete is a bit odd. The interior is well-executed, though it lacks some definition that could be improved with upscaling and inpainting.

Better than I expected in terms of realism. A small step below Midjourney, with the disadvantage of lacking generative editing tools to enhance the result.

 

It looks like a model. Not bad, but not great either.

/8/ PERSONAJE 3D

Un personaje animado en 3D, que encarna a un viajero del tiempo juguetón y aventurero, vestido con la icónica moda de los años 80
A 3D animated character, embodying a playful and adventurous time traveler, dressed in iconic 80s fashion
Midjourney 6 🟢
DALL-E 3 🟠
Adobe Firefly 2 🔴

El coche parece un poco desproporcionado pero el conjunto de la imagen está bastante conseguida. Acudiría a Midjourney si quisiese tener mucho control sobre la imagen con elementos concretos.

 

Muy guay, aunque el muñeco en sí no es de mi gusto, el resto de la composicón está muy lograda. Sigue pareciendo 3D pero un buen 3D.

 

 

 

Mal. Muy mal. Ni es estilo 3D y el delorean lo ha representado como un UFO.

 

 

 

 

 

/9/ ILLUSTRATION: STICKER ART

An urban DJ character in a minimalist 'Sticker Art' style.
prompt: A minimalist 'Sticker Art' style urban DJ character using three contrasting colors for impact. Features include oversized headphones, sunglasses, and a bomber jacket.
Midjourney 6 🟢
Dall-e 3 🟠
Adobe Firefly 2 🔴

Surprising results in illustration. Modern and urban style, which is exactly what was requested. It’s not very “sticker-like,” but it showcases Midjourney’s new capabilities for creating eye-catching illustrations.

It meets the prompt. DALL-E is a good tool for creating illustrations, although it sometimes suffers from a lack of originality.

Terrible. 

/10/ URBAN PHOTOGRAPHY

A nighttime urban street scene with an individual in a lilac dress and pink hat leaning on a barrier
prompt: A nighttime urban street scene with an individual in a lilac dress and pink hat leaning on a barrier, capturing the diverse and vibrant spirit of city life
Midjourney 6 🟢
DALL-E 3 🔴
Adobe Firefly 2 🟠

This particular image is a bit strange, but Midjourney is currently the go-to tool if you want to create photography. It’s the most realistic of the three.

You can’t expect much from DALL-E in photography; this is the best you can get.

Although the model is well-rendered (except for the arm-rail error), the lighting is strange, and the background doesn’t look realistic. Perhaps changing the lighting could yield better results, but I remain skeptical.

/+/ THE GRADES

prompt: 3D render of the number "7" in deep green engraved with white and yellow flowers on a white background.
prompt: Renderizado 3D del número "7" en verde profundo grabado con flores blancas y amarillas sobre fondo blanco
Midjourney 6
DALL-E 3
Adobe Firefly 2

bad

🔴 Nothing

Average

🟠 Friendly Interface

🟠 Text

good

🟢 Ilustration

🟢 3D

🟢 Photography

⚡ Control over output  / Original

bad

🔴 Photography

regu

🟠 Friendly Interface 

good

🟢 Ilustration

🟢 Text

🟢 3D

🟢 Control over output.

⚡Contextual comprehension

bad

🔴 Text 

🔴 Ilustration

regu

🟠 Control over output

good

🟢 Realistic product photo

🟢 Esasy Interface
⚡ Photo hyperrealistic in portraits

/// VEREDICT

Una carrera reñida entre Midjourney y DALL·E con Firefly rezagado, representada con coches de carreras en una pista futurista.
prompt: A close race between Midjourney and DALL·E with Firefly lagging behind, depicted with racing cars on a futuristic track
🏆 DALL-E 3 + Midjourney🏆

BRONZE 🥉

Firefly fails in almost everything except photography. There are no significant changes.

SILVER 🥈

Version 6 of Midjourney has significantly improved in terms of photographic realism. At times, it’s still challenging to get it to fully “understand” what you’re asking for, but thanks to its tools and commands, it’s relatively easy to maintain control over the results. Text generation has also improved, but not enough to be fully reliable. Good progress overall, though it’s still not a perfect tool.

SILVER 🥈

DALL-E remains my favorite tool because of its multimodal capabilities. It’s perfect for creating visual concepts, but it has its limitations when it comes to controlling the output. It’s not designed for generating mass content.


CONCLUSIONS  💡

  1. Midjourney (V6):

     

    • Strengths: Midjourney has made remarkable improvements in photographic realism, making it a top choice for photo-like imagery. Its tools and commands offer decent control over results, and it’s made strides in text generation.
    • Weaknesses: While it offers great realism, it can still struggle to fully “understand” some prompts. Text generation is improved but remains unreliable.
    • Conclusion: Excellent for high-quality, realistic images but not perfect yet.
  2. DALL-E:

     

    • Strengths: DALL-E remains a favorite for its multimodal capabilities, allowing it to handle both text and image generation. It’s particularly strong in creating conceptual visuals and unique, creative designs.
    • Weaknesses: Its limitations in controlling the output make it less reliable for large-scale content generation.
    • Conclusion: Perfect for conceptual and creative visuals, but not ideal for mass content creation.
  3. Firefly:

     

    • Strengths: Firefly excels in photography, offering good results in that field, and it’s great for more refined, high-quality imagery.
    • Weaknesses: It still struggles in most other areas like illustration and text generation, showing minimal improvement over time.
    • Conclusion: Best for photography but lacks the versatility and improvements needed in other creative areas.

Overall Summary:

 

  • Midjourney continues to lead in realistic image generation, especially for photography and artistic results.
  • DALL-E is best suited for conceptual and creative tasks but needs more consistency and control for large-scale generation.
  • Firefly shines in photography but is still lagging behind in other areas like illustration and text-based tasks.

Of course, this isn’t an objective analysis, nor is it meant to be. If you disagree with what’s said here, feel free to share your outraged and passionate opinions on all your social media platforms.

See you in the next ultimate comparison!

And don’t forget to subscribe to the Newsletter!

 

Unlocking Visual Creativity with AI: How LORA Models Can Boost Automotive Brand Marketing (I)

Today, the integration of human intelligence and artificial intelligence has evolved from a futuristic concept into a tangible reality. The lines between the human mind and the machine’s potential grow blurrier by the day, and it’s within this shifting landscape that OYSTERS emerges—an agency defined by an AI-First approach, placing artificial intelligence at the heart of everything we do.