I know it’s been a very long time since I’ve written anything. Almost a year considering my last post was welcoming in the new year (sorry). But today I present you with something very different to my normal content. This is the start of a new series, ‘What Midjourney thinks about my art’, or something along those lines.
For those of you who don’t know what Midjourney is, I’ll try and explain it briefly. Midjourney is a generative artificial intelligence program that creates images from natural language descriptions, called prompts. It runs via Discord which is essentially an online chatroom. If you subscribe to Midjourney you can have private conversations with the bot and Midjourney’s multiple tools. These include imagine (telling it what you want to create), blending multiple images and getting it to describe pictures for you too.
It’s that final feature that I wanted to try out for this post. I used it a few days ago to work out how best to describe a jacket we recently bought for our dog as I wanted to generate a cute picture of him wearing it. However, the prompts I was writing weren’t producing what I wanted.
This experience led to me thinking, ‘I wonder how Midjourney would describe my artwork?’. And so I bring you a blog series where I show you the original piece along with the four prompts generated by Midjourney and the art that it consequently creates.
Showing Midjourney Simon Squirrel
First I wanted to try out this image. When I drew this I was doing a daily challenge on Instagram. The prompt for that day was ‘bounty hunter’ so I took inspiration from my favourite comic book character Rocket Raccoon, more specifically the Skottie Young version.
I was intrigued to see if Midjourney would recognise this inspiration or not and what other artists it may associate with this style. It did not get the Skottie Young reference but did think it was similiar to Alex Hirsch which made me quite happy.
Midjourney’s suggested prompts
Anyway, the four prompts that Midjourney came up with were these:
- a cartoon cartoon squirrel holding tools, in the style of anime-inspired character designs, purple and amber, trashcore, joram roukes, spiky mounds, genderless, unicorncore
- an illustration of a squirrel holding a pair of pliers, in the style of unicorncore, alex hirsch, staining, raina telgemeier, spiky mounds, genderless, rough-edged 2d animation
- comic squirrel guy holding a crow axe and wrench, in the style of light magenta and amber, shige’s visual aesthetic style, appropriation artist, spiky mounds, colorful animations, caninecore, fawncore
- a squirrel with a toolbox and a shovel on his hand, in the style of anime-inspired character designs, light amber and purple, spiky mounds, genderless, loose brushwork, fawncore, unicorncore
When assigning numbers to images in the following analysis, I follow the same pattern as Midjourney, 1 and 2 are on the top row going left to right and 3 and 4 are on the bottom row accordingly.
a cartoon cartoon squirrel holding tools, in the style of anime-inspired character designs, purple and amber, trashcore, joram roukes, spiky mounds, genderless, unicorncore
Prompt one considered the illustration to be more ‘anime’ than ‘comic book’ but picked up on the vibrant colour scheme. Joram Roukes is not an artist I’ve heard of before and definitely not one I would associate with my art style. However, the artwork this prompt created was intriguing, to say the least.
Images 1 and 4 are probably the least accurate in comparison to the original. I like image 2, it has a Rocket Raccoon vibe to it. And as gruesome as image 3 is, it reminds me of some of Skottie Young’s other work only 3D. I will play around with image 2 a bit and see what variations Midjourney comes up with. He seems like a mechanic from a kid’s show which I think is pretty cool but I’d like to see it in more of a hand-drawn style.
Now I feel like prompt two is probably the least accurate. It worked out it was a squirrel and a cartoon. In fact, rough-edged 2D animation is probably a good description of the original style, however, that doesn’t translate into the actual images. And the distinct colour scheme was completely overlooked.
Why three out of four images are in dungarees and the fourth naked I don’t know. Images 1 and 3 have a very similar vibe to each other but it most certainly does not match my style at all. I like the crazy look on Image 2’s face, I can definitely see him in a comic book or cartoon show. He looks like he’s about to murder someone with his slightly distorted woodworking tools. He’s also the only one that is anthropomorphised in any way other than having articulating fingers and clothes. As for image 4, well it looks nothing like mine but could certainly be from an old-fashioned comic. Looks like something you may see in Archie. All in all, I’d say this one is a fail but a fun experiment nonetheless.
comic squirrel guy holding a crow axe and wrench, in the style of light magenta and amber, shige’s visual aesthetic style, appropriation artist, spiky mounds, colorful animations, caninecore, fawncore
This is, on all accounts, the most accurate of the four prompts. I must admit I have no idea what ‘shige’s visual aesthetic style’ but it seems to have workde. That’s what’s so fun about using existing images to create prompts, you get introduced to new styles. I mean who knew that caninecore was a thing? And there’s definiely a lot of vibrancy and movement in all four pictures.
Side note: If you’ve been paying attention to the prompts vs results you’ll have noticed in this image set, that the prompt specifically states that the character is holding two items, yet in all four pictures he’s only holding one. This is a common issue with Midjourney. I first started using this program to create an avatar for a new roleplaying character I’m working on. He’s a cat-folk thief who uses double knives. No matter how hard I tried I could not get Midjourney to give him a knife in each hand and had to rely on my Photoshop skills to do so. It also struggles with specific colour requests. I already mentioned the doggy raincoat in my intro, but the same went for giving my character ginger ears and tail. So I guess what I’m saying is, as great as it is, I don’t think artists or graphic designers need to be too worried yet. Now back to the blog.
a squirrel with a toolbox and a shovel on his hand, in the style of anime-inspired character designs, light amber and purple, spiky mounds, genderless, loose brushwork, fawncore, unicorncore
Although the prompt describes the style as anime-inspired, it seems fawncore and unicorncore took over. I genuinely put off analysing this set of images as I just find them all very boring. Cute but boring. I ranked this prompt’s results above those of number 2 solely because it managed to retain the colour scheme.
I think image one shows the ‘anime’ aspect of the prompt most with those big purple eyes, however, this also gives the character a very feminine look and the prompt did refer to this as being genderless. Number 3 looks like it’s come straight out of a 3D Gruffalo movie, just a mundane squirrel in a forest who happens to own a toolbox, or maybe he’s a Beatrix Potter character. Image 2 looks like a delivery boy with a giant spade. And I quite like 4, though it is nothing like my original image. I feel it has a lot of depth to it. You can imagine a long backstory existing behind that guy.
Running the Midjourney’s prompts with my image as reference
So another thing you can do to help generate images is use other images. So next I wanted to see what would happen if I included the original version of Simon Squirrel with the prompts Midjourney created. I won’t lie, I had originally planned on seeing if I could get a similar image based on how I’d describe the character. And adding in the original helped significantly so I wasn’t surprised by the improvement in results.
Rather than analyse each image, I decided to choose my favourite from each prompt. Which was actually much harder to do than one may expect.
Image 1 definely gives Rocket Raccoon vibes which is why I like it. The stance is also similar to my original and the balance of colour is good. Image 2, which comes fromt he prompt I originally thought was least accurate, picks up on my style very well. The small, slanted eyes, roughly sketched lines etc. He could certainly be a comic book bounty hunter. Image 3 is just so colourful. Again the line weight is really aesthetically pleasing and matches the vibe I was originally going for. There’s detail without it being overwhelming. All in all I just really like this style. And finally image 4, honestly, yet again, this prompt just created a bunch of cutesy squirrels holding indeterminably tools. I chose this image as my favourite simply because I like the feet.
So, what do you think of the images Midjourney created? Which was your favourite? Was this blog interesting to read? Did you enjoy it? If so what kind of character would you like me to analyse next? Should I go into more depth and include my attempts at creating my own prompts to regenerate my existing artwork? Feel free to answer all these questions and give any other feedback you have in the comment section below. If you do comment please do so on this post and not whatever platform you see it on (it helps with techy stuff).And as always, if you haven’t already make sure you subscribe, the form is just down below (slightly off to the side).