AI Showdown: Comparing Midjourney, DALL-E, Firefly, Meta & Stable Diffusion on Identical Prompts

A couple of days in the past, we had an early Christmas existing from the Midjourney crew with the sudden release of V6’s base model, promising far better prompt comprehension and text generation than its previous model. A week ahead of that, Meta also dropped a new AI picture generator, which I think is the ideal totally free model proper now.

So, it truly is that time of the yr once more. No, I am not speaking about the vacation season.

It truly is time for a major comparison amongst the market’s most common AI picture generators: Midjourney, DALL-E, Firefly, Stable Diffusion, and Meta.

Which a single will come out on prime this time? Spoiler Alert: The solution could not shock you.

The Greatest Output Comparison

This is the most significant comparison we have ever created, so I will use the identical prompt for every single picture to sustain fairness. I will also prominently show the ones I like the most, but never fret: I will label every single image to stay away from confusion.

Reasonable (Portraits)

shut-up portrait of a weathered fisherman, wrinkles close to his eyes,
salt-spray on his beard, hyperrealistic textures, cinematic lighting

Amongst the 5 picture generators, only Midjourney and Meta managed to generate pictures that would pass the smell check. Firefly’s portrait is as well waxy and the fisherman’s beard appears fake. Secure Diffusion does not search sensible at all, but far more like an oil painting. DALL-E 3could’ve been very good, but it overemphasizes on the wrinkles.

Seem at the particulars on Midjourney’s picture. When if you zoom in, you can see each strand of hair, the age lines, even the reflection on his eyes. It also has steady lighting and depth of area. Meta is a shut 2nd, but it nonetheless has that “softened” impact which is a trademark for AI picture generators at this stage.

Reasonable (Landscape)

a rugged coastline eroded by relentless waves,
towering cliffs which is been sculpted into dramatic arches and hidden coves,
seabirds soar over, mist swirls along the horizon, realism

When once more, Midjourney wins this round. V6 actually has been a gamechanger when it comes to sensible pictures. The pictures it outputs is nonetheless a minor stylized and vivid, but it can now pass as a genuine picture. Nonetheless, if you are just seeking for a landscape stock picture, then Firefly may be the far better choice for you.

As for the other 3: Secure Diffusion and Meta have been truly quite respectable, but the cliffs search like a lump of smooth clay when zoomed in. DALL-E three opted to make digital artwork, which is not what I am seeking for.

Reasonable (Sports activities)

freeze the action as a pickleball player scores the
ultimate stage to win the globe championship

Okay. There is a great deal to unpack right here.

Midjourney is the clear winner of this group. It completely encapsulates the fast-growing sport of Pickleball and the kinetic vitality behind it. DALL-E three could’ve been very good but it suffers from repetition of particular components.

Moving on to the bottom 3. Adobe Firefly appears to be the ideal amongst them, but it truly is not an real photograph, there is no paddle, and the player only has 3 fingers. As for Secure Diffusion, the player is not utilizing the appropriate products, he is breaking by way of the net, and his encounter is melting. Virtually.

This Meta picture however. It truly is freaking hilarious. No more comment.

Vogue

a elegant guy, in the type of orange and green, plants,
postmodern photography, shadow perform, classy figures,
artwork nouveau vogue

Midjourney appears most like genuine vogue photography, so it truly is deserving of a 1st spot for me. The only issue I have with it is that the shadows obscure elements of the outfit, which must be the emphasis in the 1st spot. Meta developed the ideal prime but it would’ve aided if we could see the whole outfit.

DALL-E three is so very good but the subject’s shadow bothers me as well significantly. Secure Diffusion has very good photography, but a rendering concern triggered the fingers to bleed into the outfit. Adobe Firefly is so sensible, but it did not comply with my directions for artwork nouveau or elegance. This would’ve been way larger if it was for informal vogue.

Architecture &amp Interior Style

a sensible dorm area, interior style,
golden hour, noisy, urban, atmospheric

In terms of realism, only Midjourney and Meta passed this interior style check. I truly choose Meta right here simply because it appears like an real dorm area. Positive, there are nonetheless some errors, largely in the laptop display on the left, but it truly is unnoticeable from afar. Midjourney’s output is very good as well, but its nuance feels off given that that is not a useful dorm area style.

3D Item Renders

industrial photography, a perfume bottle,
pastel blue background, dreamy, soft lighting, centered, flowers

I am truly impressed simply because all of these turned out to be very good. Nonetheless, Midjourney V6 continues to be on a league of its very own with an additional gorgeous entry. It truly is dreamy, nicely-shot, and has excellent contrasts. Meta is, after once more, a shut 2nd. The only letdown is the undesirable text generation.

Character Style

character style, a human battlemage, forest imagery, inspired by substantial fantasy

DALL-E 3’s artwork in this round is so extraordinary. It truly is an wonderful template if you are seeking for a smart NPC for your following DND session. Secure Diffusion tends to make far more sense if you are brainstorming a character for by yourself or a hero in a game.

Midjourney could’ve been very good, but the selection not to show the subject’s encounter tends to make completely no sense for character styles. Firefly is a minor as well mainstream for my taste and it appears like an NPC from a single of individuals outdated Adobe Flash video games. Meta also created a excellent style, but I would argue that it is not a battlemage.

Digital Artwork

pixel artwork scene, a quiet and empty supermarket at evening,
atmospheric, sixteen-bit

This is a matter of private preference, but on cautious evaluation, I choose DALL-E and Secure Diffusion’s edition of this prompt simply because it completely emulated the “atmospheric” vibe I was seeking for. This is also the 1st time that Midjourney comes at 2nd for me, mainly simply because the “pixel artwork” illusion goes away when you zoom in.

Midjourney has a very good entry, but the pixels are as well fine to the stage that I never feel it must qualify as pixel artwork any longer. Firefly did not crack the prime two simply because it created foods marketplace stalls within a grocery, which demonstrates that it lacks nuance. Meta is, by far, the worst in pixel artwork, failing in each contextual comprehending and pixel artwork impersonation.

Emblem

a emblem for a barbershop, by paul rand, clean background, minimalist

This is a win for Midjourney. Absolutely everyone else went for a generic emblem, but Midjourney did anything new by taking a barber’s pole and turning the colours into anything that resembles brush strokes. It truly is so basic however so efficient and exclusive. Apart from totally fulfilling a extended prompt, this is possibly the ideal situation for Midjourney’s enhanced nuance.

DALL-E 3 also deserves a mention right here simply because it managed to generate a nicely-created emblem, albeit typical. The most significant issue I have with it however is that it developed two distinct logos when I asked only for a single.

Text Generation

a comic panel of a distraught Tony Stark saying “Captain is dead.”

It must come as no shock that DALL-E three is in our Leading two this round, but for the 1st time ever given that I have commenced evaluating AI picture generators, I never locate it the ideal for text generation. But let us begin with the Secure Diffusion, Meta, and Firefly 1st — all of which could not create legible text. Oh, and I never feel Firefly is aware of who Tony Stark is.

When Midjourney V6 came out, they place an emphasis on their text generation enhancements and it actually demonstrates. Seem at the accuracy of that text. That is not even edited. I have mentioned it earlier in my V5 vs. V6 comparison, but Midjourney actually is the ideal at text now.

Now, let us go to DALL-E three. It could not be as very good as V6 but it truly is virtually there. Practically. It surely did not assist that Tony Stark is shouting “Captan’s dead” whilst Captain America is behind him.

Large Context

A middle-aged girl of Asian descent, her dark hair streaked with silver, seems fractured and splintered, intricately embedded inside a sea of broken porcelain. The porcelain glistens with splatter paint patterns in a harmonious mix of glossy and matte blues, greens, oranges, and reds, capturing her dance in a surreal juxtaposition of motion and stillness. Her skin tone, a light hue like the porcelain, adds an virtually mystical good quality to her kind.

This one’s truly extraordinary. If we’re only speaking about comprehension, then all of these pictures passed this check. So, we have to aspect in which a single fulfilled it the ideal.

I took this prompt from DALL-E 3’s announcement webpage so there is no query that their output is the ideal. From there, it truly is challenging to rank the other individuals one to four.

Secure Diffusion and Midjourney had the ideal seeking outputs, but it tearing does not search like “broken porcelain” to me, far more like a crumbling wallpaper. Firefly was virtually best, but it missed the “splatter paint patterns.” Meanwhile, Meta fulfilled each factor of the prompt, but it created a subpar picture, in my viewpoint.

So, What Are They Great At?

Last Ideas

With the release of Midjourney V6, it truly is acquiring tougher and tougher to make a situation for other AI picture generators. The base model is on a league of its very own, and it truly is only going to get far better when they officially release it specially given that they are taking consumer viewpoint to enhance their model.

Nonetheless, if you are just a informal consumer, Meta is a very good different given that it truly is totally free. If you are seeking for a model with wonderful comprehension, DALL-E (with ChatGPT) is nonetheless the ideal a single in the marketplace. Adobe Firefly must also be a excellent different if you are utilizing it with other Adobe goods and introducing streamlined inpainting to your workflow. Lastly, Secure Diffusion can also enhance a great deal with LoRAs and other extensions, which I am arranging to employ in long term Secure Diffusion testimonials.

The truth of the matter is that there is a great deal to enjoy for each AI picture generator out there but V6 is a genuine turning stage for AI artwork. The only query is, in which do they go from right here?