How I Crafted my AI “Style” for AI-inktober

I took a four step approach to creating my style, and I'll walk through them in detail in the following article. Those steps include gathering inspiration then categorizing the and narrow down my selections, using MidJourney to describe the images, running the descriptions to test and visualize the outputs provided, creating a mega-prompt based on all the styles of output I liked, and testing with different scenarios to ensure that the style was replicable. The journey to creating my style was one I thought would be short and sweet and turned into a days-long activity to hone in on my ideal outputs.

Step 1: Gathering Inspiration

In my journey to develop a distinctive style, I drew inspiration from the works of artists whom I deeply admire and I explored Pinterest, where I've curated a rich source of diverse artists for my feed over the years. Each visual distinctly drew upon specific attributes I was looking to harness in my final output. They didn't collectively all have to have what I was looking for, just pieces to help me assemble what I wanted to convey. By blending these various sources, I was able to collect a variety of images that spoke to my final vision.

Step 2: Describe the Images with MidJourney & Run the Descriptions

I went into MidJourney and used the describe feature to help me try to understand how MidJourney would prompt an image similar to the styles I've gathered. As MidJourney describes it this feature "generates prompts that are inspirational and suggestive, it cannot be used to recreate an uploaded image exactly". I think this is the part I enjoy the most, because a few things happen. The first, is I learn to understand how to communicate with the AI a little better. But the second, the second is seeing all the interesting and strange things that the AI interprets of what I've uploaded. From incorrectly assuming subject matter, to adding it's own unique styles, and even providing artists of inspiration whom I've never heard of before. It's a great way to just see some very unexpected visual stimulation.

What’s funny is the first image I tried, the describe tool gave me some odd interpretations, and after running all four examples it gave me, the results were some very off-base images.

Here is the image I uploaded and the descriptions MidJourney provided:

I selected the "run all" command to generate images based on all four of those image descriptions. What I got did not exactly give me anything too inspirational, so I then attempted to describe that image myself and the results were a little more expected.

Here's what I got from running the four description results.

"edgy art, a dark and mysterious animal drawing, in the style of light purple and red, bunnycore, detailed botanical illustrations, meticulous surrealism, lowbrow, precisionist lines, undefined anatomy --ar 10:13"

ali, hildys the owl the purple flowers and rabbit, in the style of marco mazzoni, neil welliver, charles vess, prehistoricore, macabre illustrations, bunnycore, commission for --ar 10:13

an illustration with a dark rabbit in the middle, in the style of light purple and crimson, detailed anatomy, nature-inspired pieces, grisaille, witchcore, made of vines, yup'ik art --ar 10:13

a doodle of a raccoon with purple flower buds, in the style of emil melmoth, eerie dreamscapes, george underwood, takato yamamoto, detailed anatomy, bunnycore, dark silver and light purple --ar 10:13

While I'm not opposed to a lot of these, they certainly don't align with the visual I created. So then I input my own prompt and got something I could pull inspiration from.

a black ink pen drawing of a rabbit that is half furry and half skeleton, the rabbit is encompassed by flower petals hand-drawn in black ink, petals filled with light and dark purple watercolors

(also notice I took off the --ar 10:13 because I wasn't trying to retain any particular aspect ratio in my final output. It put that in based on the image I uploaded using the describe feature)

You can see just from this alone, I am getting more of what the original image was in pieces, and more of the style I was looking for.

Lesson #1: Trust the way you speak and describe images you like, it'll probably turn out better than the AI. You know what it is you like about the image and what you want to draw from it.

From there I repeated that process with a few other images.

This one was the one that resonated the most: a drawing of a lady in water over her head, in the style of detailed ink, fantasy realism, monochromatic sculptor, anime-influenced, victorian-inspired illustrations, cut/ripped, intricate dotwork.

The result was something I could use to help me create my final prompt.

The outcome still not great, I want more contrast, more like it was drawn with a pen, so I first tried tweaking this a little bit at first to see if I could get more detail.

a detailed pen ink drawing of a lady in water over her head, in the style of black pen ink, fantasy realism, monochromatic high contrast, cut/ripped, intricate dotwork, intricate linework, hand-drawn aesthetic

This is pretty close but still lacking - so I took inspiration from none other than one of my biggest inspirations in web3, bigeggs, and input one of his illustrations into midjourney to ask for a description.

Certainly this could lend me some helpful verbiage.

I chose to run this description: an illustration of a man with a pipe, in the style of steampunk creatures, intricate use of hatching, ambrosius benson, john blanche, moche art, skeletal, sandalpunk.

Step 3: Creating a "Mega Prompt" 

What I've gathered from the MidJourney descriptions and attempts at my own are the following pieces:

• black ink pen drawing
• detailed ink
• high contrast monochromatic
• intricate dotwork
• intricate linework
• hand-drawn aesthetic
• intricate use of hatching
• satirical etchings (although satirical is really not needed here)
• hand-drawn in black ink

I opted to add the color portion of the prompt in later to first nail down if I could get the proper style of hand-crafted penmanship.

For my test I’m going to try to emulate a drawing of a fox floating in outer space with the following prompt:

a black in pen drawing of a highly detailed fox floating through outer space, high contrast monochromatic etching, intricate lifework, intricate footwork, in the style of intricate hatching, hand-sketched aesthetic.

Again, take notice I focus A LOT on the style and not what is in the image at this point, again I’m trying to nail down how to describe the appearance to translate to different subjects. I think for a first attempt this is pretty close

However, I’m still really missing that lifework that makes it feel hand drawn, something about the way midjourney is rendering the final images it wants to clean and crisp up the image, which makes sense.

I’m going to take a step back and work from the beginning starting with a basic prompt like “ a sketch of a fox in a notebook with a black pen “ and build my way up from there.

I repeated this process individually with a lot of different simpler prompts, I'll spare you all the details because as I mentioned in the beginning, this was a days-long expedition into figuring out how to hone in on the style and make it replicable. By the time I made it to the end, I had a collection of beautiful images to help me create my final images, below is one of the purple-centric ones that really inspired the final prompt.

Step 4: Stress Testing the Prompt

The final prompt ended up being: 

a rough sketch of {{SUBJECT MATTER}}, detailed and intricate pen hatching, thick pen marks, grunge sketch style, dark and muted purple watercolor tones fill the pen ink drawing as accents, in the style of hand drawn, textured sketching, thick linework, intricate pen hatching, intricate dotwork, monochromatic --no notebook

(Notice also the inclusion of --no notebook, which if you're unfamilar with MidJourney lingo is a "negative prompt". A negative prompt is basically telling the AI what you want it to ignore or want it to omit from your final image output.)

Stress Test 1: A thunderstorm (nature)

Stress Test 2: A Woman with Diamonds (human)

Stress Test 3: A Deer Family (animals)

Stress Test 4: A Victorian Desk (objects)

Conclusion

I was growing frustrated with the amount of times that I tried a describe, ran it and didn't get what I wanted. I was growing frustrated with the amount of times I would describe a scene in my head to match a specific rendering that I couldn't summon out of the ai. I was afraid that the little details I found inconsistent with the different shades of purple, the levels of contrast, how the hatchings are used, would all be viewed as poor craftsmanship.

Luckily, I was on a deadline. October 1st I needed to post my first picture, and I made up my mind that I wouldn't be playing catch up. This pushed me to do something I was uncomfortable with. My idea of perfection was holding me back from putting out what I'd spent hours and hours building.

If you have been following along with my AI-nktober's so far, you'll notice there are some similarities and differences between what I'm generating and I think that is okay. When Itake a step back and look at the full 31 days it's going to feel more consistent and cohesive than I think, and it'll be full of wonderfully creative images that I never would have gotten if I didn't let myself be okay with the fact that my original idea wasn't going to work out and that sometimes the unexpected imperfections are what make it good.