As an artist I am not a huge fan of AI art. But me not being a fan of it is not going to make it go away. That’s for sure. I have little interest in sitting here, writing prompts, and seeing what the machine can come up with. Yet I also think I have to find ways to use AI art to my benefit or I’ll get left behind. So that’s what I did this week.
I wrote recently about the one hundred and twenty faces I was drawing for my Great Gatsby illustrated project. I finally fished them and that took some time. I would estimate that it took me about two hundred hours to draw all those faces. That’s with me being able to leverage cartoon art faces I had already drawn over the years that cut down the drawing time quite a bit.
I was happy to have finished the faces and I took the time to try and work them into the overall book design but I faced a problem. All of my other illustrations were in color but these faces were black and white ink drawings. They looked a little out of place and I wanted to color them. But after coming off two hundred hours of drawing them over a month’s time I didn’t want to put another two hundred hours into coloring them. That’s what it would take. So I set about looking for a way to use AI to automate the process.
The drawings were only going to be printed about three inches high. I didn’t want to color to be complicated but even keeping it simple would take me a couple of hundred hours. So I was looking at AI apps on my iPad, I often check them out, to see if any of them would be of use. I found one called Photoleap. It promised to turn photos or simple drawings into anime characters. Color anime drawings.
I tested out a few drawings and it seemed to be okay. There were even about twenty different anime styles to choose from. It would take my strange black and white cartoon drawing and render it in color as an anime drawing. It was clearly using preexisting eyes, noses, hair, and other facial features and putting them together in a form that resembled my drawing.
The problem with the whole process is that these anime drawings were clearly not my work. Nothing of my cartoon drawing remained in them except the basic shape. It was also a little disheartening that the anime drawings were actually pretty good. I would guess that your average person would probably like the anime drawings better than my black and white cartoons. I found that depressing. But I had to get over that and press on.
My next step was to bring the AI anime drawing into a Photoshop file with my cartoon drawing. I did that and they matched up well. I set the black line drawing on “Multiply” so that the color drawing would show through beneath it. Then the problem was that the anime drawing was too dominant and over all it didn’t look like my work anymore. So I set about to change that.
The first thing I did was to turn the opacity down on the anime drawing to 80%. This make it less prominent. But it was still too prominent. I selected the anime drawing’s background colors and blurred them. That turned down its strength nicely but overall the anime drawing was still too strong. So I ended up going into the black channel of the anime drawing and turning it all the way down thus eliminating its black line. There was still a dark grey line there but it seemed weak enough.
I ended up liking the way the drawing looked so I automated the process. I made a Photoshop action (a macro) where I could open the files of the anime drawing and my cartoon drawing, hit a single button, and it would join the two together just as I described. That saved me a lot of time. But I still wasn’t done.
I had three problems with the combined drawings. The eyes, noses, and mouths. Being that the app was turning my drawings into anime ones the eyes were all too big, the noses too pointy, and the mouths too small and pouty.
The nose was the easiest one. Since the app made all the noses pointy upturned noses I had to eliminate all the nostrils. The app uses a dark shape for the nostrils so I had to select that dark shape and then use Photoshop’s content aware fill to eliminate them. That type of fill looks at the pixels around where I have selected and tries to guess what they should be in the space I have selected. It works pretty well when my selection is just a brown spot.
The eyes were done about the same way. I had to select them and then shrink them down to the size they were in my drawings. That usually left a cut out white spot in the place where they used to be so I had to select that spot and use content aware fill on them too. That was also fairly easy. Content aware fill saves a lot of time.
The app also added long eyelashes on every character. That worked fine for the women but I didn’t always like the look on the men. I ended up solving that problem just by blurring the eyes. I’d select both eyes and then use the Gaussian Blur filter on them. That made them fuzzier and the lashes less prominent. That was also fairly quick.
The last thing was the mouths and all their pouty lips. Sometimes I would just put a solid pink/red color over the lips and be done with it but other times I used the scale and warp functions to fit the pouty lips into where I drew the lips. They I used content aware fill to fix the areas that were cut out and white was left behind. This took a minute but still went quickly.
I had to tweak other individual things here and there but in general it was these three things that I had to do for all the drawings. I eliminated enough of the anime look of the color so that they now look like my drawings again. But I still had one problem. Overall the color schemes were terrible. Most of the flesh tone of the faces was fine but the hair, clothes, and background colors were not great. I would never had colored them like that. They really looked awful in general as group.
I fooled around with them a bit and ended up solving the problem by eliminating the color. I turned the color into sepia tone. I figured that would fit in with the 1920’s Great Gatsby theme anyway. All these sepia toned cartoons of the partygoers sets a mood. I think I might use them as end papers.
In the end I think I got all the coloring done in under twenty hours. It amazed me how much of the process I automated with actions (macros) and content aware fill. Plus it would have taken me ten times as long without the AI anime drawings. I obscured them enough so that without me writing this no one would be able to tell that’s how I did it but I still have my reservations about AI Art. After all this same App can take any photo of a person and turn it into an anime drawing in seconds. That’s scary for people who took the time to learn how to draw like that.