“I’m working on a graphic novel using Midjourney and here’s what I’ve got so far,” wrote user KrisKashtanova on August 15 in the #show-and-tell channel on the official Midjourney Discord server. They probably didn’t know then that within a very short time it would become one of the most talked about AI projects at the moment. The thread exploded, numerous community members wrote how excited they were about her first few pages. I contacted Kristina and tried to find out a bit more about her first graphic novel called “Zarya of the Dawn“.
Kristina Kashtanova has a background in software engineering with ten years of experience in the computer graphics field. After working as a programmer, they changed their profession to become a journalist and a professional photographer. At the moment, they do UX and Motion Design.
“AI felt like a dream come true. Being about to tell every story through images, I decided to tell a tiny story I was developing using 3D with Cinema4D during the pandemic. It’s about someone visiting different worlds that help her understand her emotions and feelings.”
“I tried to tell this story through 3D in 2021, but it was time-consuming and hard for me to do it in my limited free time. AI was different. I could create a page of a graphic novel in an hour at the end of my busy day; it was relaxing and entertaining.” To assemble the individual images into comic panels and add text, they used Comic Life 3.
“I chose a story about different worlds that starts in the future of New York City. I live in New York, so it was fun to visualize my favorite locations here with a sci-fi twist.”
AI felt like a dream come true.
“What I learned is that Midjouney connects a lot of people. In 10 days I was creating the graphic novel, I met around 1,000 people who were inspired to share their own stories through this medium.“
People are especially excited about their achievement, essential for a graphic novel, of generating consistent faces. “As for the characters, it was hard. I used a celebrity at first as the easiest way to achieve consistency, but then I learned how to create my unique characters with MetaHuman and bring those to Midjourney.”
“Some of you mentioned that you didn’t feel good that I used Zendaya without her permission, which made me think”, Kristina explained on their Instagram page. They finished their first draft using her as my main character, but in their second draft, they replaced her with a character they generated themselves. “I chose her because of her activism work that I’m deeply inspired by and because she’s a Person of Color and MJ didn’t work too well with non-white characters. I didn’t know any other way to make a consistent character rather than using a celebrity. Now I know and I promise to fix this issue in my next draft (along with other helpful feedback you generously provided).”
Kristina dedicate the first character to their grandmother, “who raised me and was my best friend for 35 years of my life. She passed away last year. Her name was Raya.”
“I fed screenshots of this MetaHuman to MJ and generated color pencil sketches of different expressions. MJ has trouble working with expressions, but MetaHuman allows to generate a character with expressions that can be used as a source image.”
“There’re limitations that Midjourney has at the moment. It doesn’t do well with multiple people in the frame and any kind of action, which is tricky for a graphic novel. There’re methods to go around it, but it is not guaranteed and would take a lot of time. The challenge and limitations motivated me to do it because I like solving problems.“
“Midjouney has a history; depending on what you visualize and which words you use it’ll generate images a certain way; that’s why I feel my environments and style were consistent. As a photographer, I would try to prompt my images as I would direct my scene during a photoshoot.”
Kristina is not the only person who came up with this idea. In fact, they were inspired by another artist. Elvis Deane is also making progress in creating his first graphic novel using Midjourney and recently shared his experience in a video.
Elvis has more than 20 years of experience drawing comics and was blown away by the possibilities of Midjourney right from the start. After spending his first free credits within an hour, he quickly had to upgrade to Basic and then Standard Membership. However, he also quickly encountered the problem of producing uniform faces and used the names of well-known actors for this purpose.
“So within those first few hundred rounds of images I realized, I was sort of developing the story. I had this main character, I had a dream, that she was gonna be having, and some sort of elements of what I want this dream to be. I started to just iterate from there and try her in different situations with these goat characters and see what came up. Within a couple of hours I had finished a comic I had written, I had laid it out in Comic Life 3 and honestly, it was one of the most amazing creative days I’ve ever had in my life. It was just so satisfying to do that.”
“But then I had 10 pages of this sort of dream sequence comic that ended with her waking up from the dream. I had more of a story I wanted to tell, I had probably 10 more pages worth of images that I had developed for the further part of the story. But it was time to go to sleep and take a break.”
AI is like doing improv with the most brillant actor who doesn’t understand the language properly you are talking.
Over the next 2 weeks, he slowly started to generate more images and test things out to see how he could tell the rest of the story that he wanted to tell. “There is a lot of animal analogies that I could bring up, like trying to talk to your dogs, because man, I tried to talk to my dogs, and they have no idea what I’m doing. It’s a little bit like that with an AI. It is also like having an infinite number of monkey artists that are doing something and it’s sort of like what you want but not really.”
“I think the analogy that is closest to how this whole thing of creating this goats comic is: Doing improv. If you have ever done any improv, it’s saying ‘yes and’ to whatever your scene partner is doing. So you are sort of building on what they say and they are building on what you say and together you are telling a story. This was a lot like that except it’s like working with the most brillant actor who doesn’t understand properly the language that you say to them.”
“I would say I want this character to be in a book shop in front of a shelf of books. Sometimes her head would be in the shelf of books, sometimes her head would be floating somewhere, sometimes 2 heads would appear in the same shot. You never really had predictable results. You could feed what you thought is a perfect instruction, but it just didn’t come out the way you wanted. I found myself rewriting a lot of what I wanted to do, where I had to write bridging information. I would look at a panel and say this is what I would ideally like it to be, but I’m gonna have something else put there and maybe write a caption to sort of explain that bridging.”
“The comic ended up being a lot of sort of like a personal journal or her memoir of the situation. A lot more telling than showing and as a storyteller that’s something I was told never to do. It seems to have worked in this story. I think it came out pretty good. I’m pretty happy with the results and I have gotten a lot of positive feedback so far. But it was a very imprecise tool.“
“One of the most difficult things was action. I wanted to have an action sequence of this one character being killed and the main character running off from the villain. And it is almost impossible to get Midjourney to do any kind of action. I’m not sure that this is specifically because they banned certain words, because I wasn’t using any of the banned words. Anytime I typed in a ‘run’ command, maybe one of her legs would be off the ground. A lot of times she was just kind of standing in an alley. It was very frustrating to try to get that action but eventually I got things that kind of worked. The hardest thing I think I had was trying to get her to duck or just hide behind a wall. I could never get a satisfying image of somebody going sideways.”
“One of the best examples of how an AI doesn’t understand what a human might understand is: I typed that I wanted the character that she ducks behind a wall and the image generated had a bunch of vague ducks next to her rather than her ducking.”
Elvis was so pleased with his graphic novel, which he created in such a short time with Midjourney, that he also shared it online in the comics subreddit. As a result, he received a lot of positive feedback – but also what he thought were legitimate discussions. One young comic book writer, for example, wondered what this would mean for the security of his profession in the future. “I don’t know that I can answer that. I think right now any artist is kind of safe in able to create things much more precise and much more specific than an AI can generate. Will that continue? I mean this technology seems to be. Midjourney has been only around 3 or 4 months. It’s growing really exponentially. It’s incredible how much V3 has changed since V2 and how much more precision it’s having in creating faces and bodies.”
There is also the matter of copyright, which is much discussed in the context of AI image generators. According to Midjourney’s ToS, the person who entered the prompt owns the resulting image. But what about the text in the prompt itself? At the same time, there are now even the first marketplaces where these text inputs are traded. In the case of Elvis, the fact that he used well-known actors as references is particularly piquant – do they possibly also have rights to the images? Theoretically, they could have their images removed from the databases used to train AIs like Midjourney. Elvis compares it to deepfakes, which people would also already monetize in some cases.
Over the past few years he has created 100 one-panelers, a few short stories, mostly jokes, but never a longer series that he was satisfied with. In the near future he plans to finish the story section with two more issues. “To have this tool to be able to do that I think is wonderful. I do see it even in my comedy funny comics that maybe I’ll draw all the characters but use AI to generate the backgrounds. There is a lot of debate within the AI art community, almost every day somebody says ‘but you are not artists’, and I agree with that.
“I really see this as an exercise in art curation, art direction, creative direction. So many people have ideas in their heads that they have never been able to get out in a satisfying way and this is getting them far closer than they ever had a chance before. It’s beautiful and it’s exciting and it is also a little bit terrifying because we don’t know where it’s gonna go. There is just so much that they can do very early on and it’s a little bit unnerving but I don’t think you could put this genie back in the bottle and we just kind of have to figure it out as creative people, what we are gonna do with it and how we are gonna use it ’cause yeah, it’s here. And it’s scary and it’s going to eat your brain.”
In the meantime Kristina has finished their first issue of “Zarya of the Dawn”. It took them 12 days and about 1,500 prompts to complete the 17 pages. A great news that underlines how wonderful the community of the AI image world is: Their graphic novel will be translated into Bengali, Hindi and French and thus be made accessible to an even larger group of people. Could they imagine holding a printed edition in their hands at some point? “I’m not planning to print it or to monetize on it in any way because I made it to share my story, to connect with others, have fun and learn AI. I’ve been only using AI for 15 days and it’s my first graphic novel. I have a lot to learn. It was incredible fun and rewarding though.”
Recently, Midjourney had released for just one day a new beta of its AI based on Stable Diffusion, which opens up whole new possibilities for a comic artist, especially with its impressive results in generating faces. “I felt that it was impressive. They let people use it without any limitatons for several hours and I’m sure it won’t happen again. Like, DALL-E 2 there’ll be some sort of censorship. I noticed that a lot of people didn’t feel at ease with such photorealistic images. But it gave much more freedom of expression. Besides, people can still use version 3 if they want to.”
“As for the second issue I’d like to wait until the new version of MJ comes out. I feel like I already learnt everything I could and experimented enough with version 3. When the new version comes I’m thinking of making a second issue to learn and experiment with it and share with others my process.”
“My goal was to learn AI. I made this issue in 12 evenings, never using any AI platform. I use it as a tool to sketch my ideas”, she adds on Instagram. If I ever plan to publish this, I would hire a real-life artist to re-draw it. Due to the current big limitation of Midjourney platform, I built my main character based on Zendaya (as I respect her work and activism and look up to her as a human). In my final graphic novel, of course, it would be a different character created by the artist.”
“I created two other characters from scratch using MetaHuman. Unfortunately, MetaHuman characters are limited and would be hard to use throughout the story, otherwise I would re-create my main character. I hope in version 4 of MJ it’ll be easier to make (and I believe it will be because of the deterministic nature of Stable Diffusion algorithm).
I find it incredibly fascinating how quickly artists are discovering new tools, mastering them and creating things that would not have been possible in this form just a short time ago. I thank Kris and Elvis from the bottom of my heart for sharing so much about their experience and hope it can inspire many more people to let loose artistically! If you have any works you’d like to share, feel free to drop me an email, comment, or on Twitter!
[…] The first AI generated graphic novels are here […]