Midjourney v5 vs. DALL-E 2: Which AI Is Better at Generating Hands?
AI art generators continue to impress, allowing us to create just about anything we can imagine. However, the tech seems to have hit a brick wall when it comes to generating realistic-looking hands.
Here, we look at two of the leading generative art apps and pit them head-to-head—or hand-to-hand—to see which can generate better hands, Midjourney v5 or Dall-E 2. Have either one of these apps mastered hands? Let’s find out!
AI’s Problem With Hands
Since AI-generated art became widespread on the internet, there has been criticism regarding thequality of hands drawn by AI. Despite recent updates, as shown in our side-by-side comparisons, the results have not been satisfactory.
Both contenders have been improving their capabilities and the quality of their outputs with each iteration. The latest update,Version 5 of Midjourney, has shown impressive progress. However, the problem with the hands drawn by AI remains unresolved and can’t be ignored.
Comparison 1: Using the Prompts “Hand” and “Hands”
Our comparisons are going to contain the exact same prompts for both Dall-E 2 and Midjourney v5. We’ll design the prompts to be hand-specific instead of simply creating people to see how the hands look. We’re also giving each app only one chance (roll) for every prompt.
Let’s start things off with the most basic and relevant prompt: “hand”.
Midjourney v5:
We’re not off to a good start!
Midjourney took the unusual route of associating a hand with rather creative situations. Instead of focusing on just a hand, we see a wizard, gloves, a skeleton, and a tiny figurine. The gloves image is missing a finger too.
Dall-E 2 takes the opposite approach and offers us just a single hand against a plain background. But oddly, there are some strange postures, particularly with the thumbs, that don’t look natural or comfortable. Each hand is also cut off at one of the sides of the images.
Who wins this round? We’ll give it to Dall-E 2 for overall accuracy.
Now, let’s make the prompt plural, “hands”, and see what the AIs come up with.
Midjourney’s attempt at “hands” turns out better this time around. But all four images are in black and white, and we have some missing fingers. Upon closer inspection, you can also see that some of the digits are strangely shaped or morph into one another.
Dall-E 2 continues to feature hands with a plain background. There are no missing digits, but the hands are cropped out in the third image and the other versions seem sort of clumsy in composition and lack creativity altogether.
Let’s call this comparison a draw. Dall-E 2 would win for accuracy if that were the only factor, but Midjourney manages to create some beautiful imagery in its black-and-white renditions, even if all four versions aren’t very realistic.
You can alwaysuse Photoshop to fix your Midjouney art, including hands.
Comparison 2: Hand Gestures
Let’s compare a couple of hand gestures that are nearly universally recognized.
Fingers Crossed
First, let’s try “fingers crossed”.
It’s safe to say that Midjourney completely botched this prompt. We’re missing fingers and none of the versions look natural at all.
Hats off once again to Dall-E 2 for getting the finger count correct, but that’s the only good news. Each version looks like the fingers are striking their own yoga poses in a game of Twister.
There’s no winner in this comparison.
Next, we’re going with “thumbs up”.
Midjourney gets the finger count correct while treating each prompt in a creative fashion. Notice the introduction of an illustrative style?
Dall-E 2 also gets points for accuracy while not trying to rock the boat with anything creative added to each result.
There’s no clear winner here.
Comparison 3: Hands With Objects
Now, we’ll up the complexity by prompting hands to interact with objects.
Hand Holding Crystal Ball
Let’s start with a random object, using the prompt “hand holding crystal ball”.
As we up the complexity, Midjourney starts to shine. Apart from a couple of the renditions looking unnatural, the hands and the crystal balls look beautiful. Midjourney even takes the time to create reflections in the glass that certainly add to the overall creativity.
But for the first time, we see Dall-E 2 missing a digit in at least one of the hands, with the fourth hand looking just plain weird. The crystal balls also don’t look as impressive compared to Midjourney’s.
Midjourney gets its first win.
Hand Holding Water
Let’s try something even more complex with the prompt “hand holding water”.
Midjourney only manages to roll one image with the correct number of digits. Though beautifully rendered, once again we start to see the cracks in the believability department.
Dall-E 2 struggles with achieving natural hands as well but does a much better job. It switches up the color in the background too for some variety.
We’ll give this round to Dall-E 2.
Comparison 4: Working Hands
For this comparison, we’ll create prompts that have the hands involved in activities.
Hands Molding Clay
Let’s see how the AI models fare with “hands molding clay”.
Midjourney missed a finger in two images but everything else looks great.
Dall-E 2’s images look confusing and crowded, resorting to adding another person’s hands in half of the versions.
The edge goes to Midjourney.
Hands Pressing Dough
Let’s try a similar activity, “hands pressing dough”.
Midjourney’s images look great overall. But once again, half of them have missing digits. But the images can’t be faulted for their artistic styling.
Dall-E 2’s versions are missing fingers as well in half of the renditions and even add one to the last hand in the set.
Let’s call this one a draw.
If you’d like to try these comparisons for yourself, we show youhow to use Midjourney to create AI art.
What Do the Results Tell Us?
It’s fun to go do comparisons and determine an overall winner. And if we had to choose, we’d call it in favor of Midjourney v5. Although Dall-E 2 created hands with the correct number of fingers more often, it was Midjourney that crafted more artistically-rendered and appealing images.
But both apps have a place in the marketplace for artists who repurpose Dall-E 2 and Midjourney images for their work. Both are capable of creating hands that can be used as cutouts or in composites for artistic, editorial, and commercial usage. It’s just a matter of personal preference.
AI Will Eventually Conquer Hands
Generative art apps like Dall-E 2 and Midjourney have come a long way in their ability to create realistic and fantastic art. They still struggle with generating hands, but given the acceleration of generative tech, we can only expect improvement in the near future.
Adobe’s vector recoloring tool will save you a lot of time and effort when you want to change your vector design’s colors.
Free AI tools are legitimately powerful; you just need to know how to stack them.
So much time invested, and for what?
Anyone with more than a passing interest in motorsports must see these films.
Quality apps that don’t cost anything.
You can’t call this offline, Notion.
