DALL-E. Meta. Firefly. Steady Diffusion. Like it or not, it truly is undeniable that the AI picture generation marketplace is certainly oversaturated now. Nevertheless, there has usually been one particular standout.
To me, it truly is apparent that Midjourney was the ideal AI picture generator in the organization. Nevertheless, I do acknowledge that it nevertheless has some flaws, especially with creating practical pictures and ones with lengthy prompts and text.
That is why I have been patiently waiting for Midjourney V6, and final evening, it lastly came. I swiftly hopped on Discord and commenced creating as numerous pictures as I could. Allow me inform you a swift spoiler: it truly is well worth the wait.
Right here are some of the ideal pictures I produced utilizing Midjourney V6 along with the identical prompt but utilized to Midjourney five.two:
Midjourney v5 and v6 Output Comparison
It is been a tiny more than 24 hrs because Midjourney v6 came out and allow me inform you: the hype is true. This has been, by far, my favored picture generator. It by some means fixed each single one particular of my issues with the prior edition. Right here are some of my favored examples:
Portraits
a female lying in bed with her eyes closed, golden hour, closeup
My most significant gripe with Midjourney was that it could not genuinely make practical pictures on par with DALL-E or Meta. The release of V6 appears to have solved that issue. Their realism is on a entire diverse degree now. No a lot more waxy faces and exaggerated characteristics. V6’s output is so very good that, even if you zoom in, you can see the imperfections that make us human. This is an immense improvement.
Landscapes
landscape, an autumn in the lake in the course of dusk, tranquility
Will not get me incorrect: V5.2’s picture is rather very good, but it truly is not precisely the appear I am going for. I am hunting for practical lake pictures, some thing that V6 was ready to give me. This upgraded edition can generate genuine-hunting pictures without having sacrificing artistic high quality. It is way greater than DALL-E three on this front, in my viewpoint.
Merchandise Renders
item photography, a perfume, studio lighting, shadow perform, jasmine, soft
I will admit: I am not as well positive about this one particular. The important big difference is that the item pictures I am acquiring from V5.two seems processed and marketplace-prepared, whereas V6 seems a lot more raw, like it truly is taken straight out of a camera. It might have some thing to do with the phrasing of my prompts because I have gotten employed to cluttered V5 prompts, some thing that I need to have to perform on due to V6’s evolved nuance.
I will say this although: if you are a seasoned editor hunting for in depth, effectively-shot raw pictures, V6 is a whole lot greater than V5.two.
Film Stills (Animated)
animated film nevertheless, a youthful woman following a magical cat to a tree,
inspired by hayao miyazaki, whimsical, magical realism, clean lines, in depth, 8k
This is a wonderful time to speak about nuance. In my prompt, I particularly requested a movie nevertheless that seems like Hayao Miyazaki’s perform. V5.2’s output did not stick to this at all, as an alternative going for a generic 3D DreamWorks type of animation. On the other hand, V6 followed this instruction to a tee. It seems straight out of Howl’s Moving Castle.
I also extremely propose you to zoom in and appear at these particulars in V6’s output. The nevertheless is so a lot a lot more vivid and total of existence. It is genuinely mindblowing how very good Midjourney has enhanced more than the final couple of months.
Film Stills (Reside Action)
movie nevertheless, back shot of a guy in a green jacket, symmetrical,
muted colours, directed by wes anderson
Midjourney V5 certainly had a issue with oversimplifying or overcomplicating prompts, specifically ones with plenty of context. Seem at the instance over: I stored it minimum but nevertheless, V5 was not ready to be imaginative with the prompt he is provided. V6 solves this issue by filling in the gaps of my prompt whilst retaining its authentic imagined.
PS. Yes, I know. The man is missing his correct ear but hey, it truly is V6’s very first week!
Flat Illustrations
brand for a shoe firm, clean background, paul rand
I by no means genuinely had any concern with creating logos with V5.two but, right after seeing these pictures side-by-side, I could genuinely inform that there was space for improvement in hindsight. V6’s output retains the minimalism of V5.two whilst including its distinctive spin to the illustrations that provides them a lot more identity.
Surrealism
the planets in the galaxy as hatching eggs of lovecraftian entities,
surrealism, cosmic, lovecraftian, ethereal, celestial bodies
I have usually praised Midjourney’s surrealist pictures as one particular of their sturdy factors. Nevertheless, it has a tendency to overpopulate its outputs with topics that you at times can not figure out what is going on — some thing that you can see over.
V6, with its enhanced nuance, manages to strike a stability amongst fulfilling the prompt and getting imaginative. You can now obviously see what they are attempting to portray, even with tiny to no details about the topic.
Text Generation
a restaurant in a quiet chic community with a neon signal that says “Closed”,
evening, streetlights
A single of Midjourney’s most significant guarantees ahead of V6 came out was that it truly is going to correct its text generation, which is a large issue across all AI generators. The only one particular I have attempted which is good on that finish is DALL-E three, but it seems like Midjourney’s following in line.
It flawlessly wrote “Closed” in the V6 picture, even including its very own flair. As for V5.two, effectively, except if you have received a restaurant referred to as “CORSTARB,” I do not believe it truly is minimize out for text generation.
Nevertheless, it truly is nevertheless not ideal, as you can see right here:
comic panel, panicked captain america yelling “Get out of right here”, speech bubbles, gritty
This just displays that Midjourney nevertheless does not acknowledge letters because it truly is nevertheless missing a word from my prompt. In my viewpoint, this functions ideal with single or two-word texts only. But hey, it truly is miles greater than its rivals. Even DALL-E three is not this very good.
Large Context
A in depth oil painting of an previous sea captain, steering his ship by means of a storm.
Saltwater is splashing towards his weathered encounter, determination in his eyes.
Twirling malevolent clouds are noticed over and stern waves threaten to submerge
the ship whilst seagulls dive and twirl by means of the chaotic landscape.
Thunder and lights embark in the distance,
illuminating the scene with an eerie green glow
Just a heads-up, I borrowed this prompt from OpenAI’s DALL-E three webpage. It is at times difficult to believe of factors to include to a prompt. This is also a prompt that OpenAI employed to check DALL-E’s nuance, so I could also check it with V5 and V6, and then assess.
V5.two in fact did a rather very good task, but nevertheless missed a couple of factors like the eerie green glow, seagulls, and thunder. V6 followed every little thing except seagulls, but there is nevertheless one particular solitary seagull in the background, so this one particular passes the smell check.
So, Did It Boost?
It did increase, by a whole lot.
I could not demonstrate you each check I have accomplished however (I am reserving some for my following write-up) but it truly is previously a hundred instances greater than V5.two in my guide. It managed to remedy the text generation and nuance concerns whilst concurrently strengthening its creativity. Every single picture I have produced so far with V6 is crisp, in depth, and exact.
What else is there to inquire for?
The Bottom Line
When V5 came out, some explained that it was a backward stage from V4.
Progressively, the staff listened to the neighborhood and enhanced its creativity, even including some functionalities in the method. The consequence was Midjourney V5.two, which was previously my favored AI picture generator in the marketplace.
Midjourney V6 is a important improvement on V5.two. It took every little thing that was previously very good with V5.two and substantially tweaked its model to generate a lot more in depth and exact pictures. Every thing that I have complained about with V5.two — nuance, text, realism — they’ve fixed that and then some.
The ideal issue is that we can only assume it to get greater from right here on out. The Midjourney staff is previously crowdsourcing consumer picture opinions by means of A/B testing to increase its model.
Mark my phrases: Midjourney V6 is a turning level in AI picture generation. Overnight tons of graphic designer jobs might’ve just been wiped out. This might’ve been the very first real realization of that.