Search online for an AI image generator and both of these will be on page one, but while you might be able to get free trials, you’ll need to pay for full access to generate more than just a few images.
Knowing which one is worth your cash investment is crucial because you don’t want to end up with two subscriptions, so we put Midkourney and Stable Diffusion in a head-to-head showdown to confirm which is the best of the pair.
The tests
Firstly, credit to Midjourney for producing double the amount of pictures that Stable Diffusion does – four, not two – after a single prompt.
However, Stable Diffusion’s user interface includes plenty more customization before you even get to the image, all helping to get you a more accurate result quickly. You can add negative prompts (details of what you don’t want to include in an image) styles, and more.
You also have control over the aspect ratio and, if you’re a paying customer, you can match Midjourney’s four-image output.
In our test, we included five distinct prompts: a familiar fictional character doing an expected action; another expected action this time being performed by real humans; a slightly more obscure prompt with a standout adjective to steer the AI, an animal doing an unlikely action; and a broader prompt that gives the AI more freedom.
“Create an image of Santa wrapping presents”
Befitting of the season, we put AI to the task of creating images of Santa doing what he does best – preparing presents.
The time it took to generate the output was fairly similar, but all four of Midjourney’s options were considerably better than Stable Diffusion’s two.
Midjourney thought about adding further context to the backgrounds and took the more traditional cartoon-like style, whereas Stable Diffusion lacked visual flair. Its Santas weren’t very convincing either – they looked like someone dressed up in a costume.
Midjourney 19.51s
Stable Diffusion 17.51s
Midjourney 1 – Stable Diffusion 0
“Depict a family walking through a field”
Both produced two utterly different styles, we were pleased with the outputs of both image generators, although it took Stable Diffusion around a third longer despite only having to make half the images.
As a Brit, the content feels very familiar. The tractor path, waist-height wheat crops and an overcast sky.
Midjourney’s output gives a warmer, Europen vibe that’s just as accurate and detailed. Its four options are all in a painted style, but both systems gave a solid effort and, with further prompts, styles can be refined to your taste.
Midjourney 18.91s
Stable Diffusion 24.07
Midjourney 1 – Stable Diffusion 1
“Create an elaborate portrait of an 1800s Queen”
On the face of it, the two AI generators nailed the brief. They’re just what we asked for, but the aesthetic of Midjourney’s images, complete with extravagant clothing and intricate backgrounds, is more akin to the type of art we see from that era. Shame that its AI-generated models are wearing a lot of modern make-up – not very 1800s.
Stable Diffusion’s output has the Queen looking much more 1800s, but it lacks creative flair and feels a bit boring.
Midjourney just about takes the point, but it doesn’t score full marks.
Midjourney 19.83s
Stable Diffusion 16.15s
Midjourney 1 – Stable Diffusion 0
“Make a lifelike image of a monkey riding a scooter”
Interestingly, both systems chose a particular type of monkey and stuck with it – neither chose the same species.
Credit where credit’s due, Stable Diffusion included background context in its images, and they’re pretty detailed.
Midjourney added more detail on the monkeys, helping with the lifelike appearance, but three of the four images had no backgrounds. I’m sure, with a more detailed prompt, this will have been included. With this in mind, Midjourney’s monkeys take our vote, and they were quicker to generate, too.
Midjourney 18.32s
Stable Diffusion 20.56s
Midjourney 1 – Stable Diffusion 0
“Show me what it would look like to be scuba diving with a blue whale”
This created the biggest time difference of all the tests, with Midjourney taking around 50% extra time to generate its images. At least it produced double the quantity that Stable Diffusion did.
All images feature the two key components – a scuba diver, and a blue whale. Proportions and perspective are also strong across the board.
Considerable it might be, we’ll put the time difference aside and call this a draw, because all six images produced represent a good effort.
Midjourney 29.68s
Stable Diffusion 19.71
Midjourney 1 – Stable Diffusion 1
Stable Diffusion vs. Midjourney: Which is best?
Midjourney 5 – Stable Diffusion 3
On the whole, there’s not a lot that separates Stable Diffusion from Midjourney, both in terms of how long it takes to generate an image and the quality of its output.
We preferred Midjourney’s images more of the time, but Midjourney’s not one to overlook, because if you have very specific visions in mind, you can control the output more before you even hit the generate button.
https://cdn.mos.cms.futurecdn.net/mUudn3sSnTkpNAi9vadSh9-1200-80.jpg
Source link
