How I Make AI Comics: Could Gesture-First Be the Next Greyscale-First?

In this post, I describe how I combine generative AI with my own pencil and iPad drawings to make the comics for my personal projects (the one that prompted this post being this one), and some musings on what counts as an artist’s own work.

I’m sort of aware that I'm wading into a touchy subject, since I never had any formal training nor was a serious member of the art community, just someone who liked to draw despite not being a talented artist, so I don’t claim to speak for artists.

I also have separate thoughts on how I'd want to build better tooling to make my process easier, e.g. I'd love to put together some sort of character library where I can keep track of my reference materials for each character, so it's easier to enforce consistency across panels and comics.

My Process

In short, for each character, I produce three things:

1. A sheet of facial expressions
2. One fully rendered example of the character, done in greyscale solely to establish light/dark values
   • I usually don't render this myself. I draw uncolored lineart, then I ask AI to render it
     • sometimes I make an honest attempt at shading my lineart in greyscale first, then ask AI to fix the parts that look off to me
3. Color palettes to apply to my greyscale rendering from (2)

Then when it comes to making the actual comics, I produce...

4. Panel-by-panel gesture drawings showing pose and line of action
   • text prompts as hint of which character is which in each panel, e.g. "character on the left is $NAME_OF_CHARACTER"
     • text prompts specifying which expression from that character's respective sheet to try, alongside any adjustments
       • sometimes memes from pop culture or photos of me myself and I doing an expression if I didn't draw that expression in my expressions sheet
   • if the characters interact with some more complicated props in the scene, like vending machines or cars or stuffed animals etc., I stub them out as cubes and such.

I write a summary of what the AI should expect to see in my panel-by-panel gesture drawings, upload the character-specific ingredients (expressions sheet, greyscale render, and color palettes), then I prompt and re-prompt until I get the results I want.

This is basically the process I would've used to draw comics without AI, borrowed from advice I followed back in my days of drawing without it.

Is Greyscale-First... "Allowed"? Is It Still Your Own Work?

Now I must say that I'm not a skilled artist. I can only confidently say I'm okay at (4) gesture drawings, (1) facial expressions, and (3) picking color palettes.

I struggle the most with (2), in fully rendering the character with convincing light and shadow.

I think that being weak at one step of the process doesn't have to stop you from making art, as long as you can break the work into steps and put your skill into the steps that you're good at.

For example, some well-trained artists do steps (2) and (3) in one sitting, i.e. they never render in greyscale first and apply color afterwards. Instead, they render directly in color because they're skilled enough to judge light and color at the same time.

Personally, though, the advice to render in greyscale first and then put color on top makes things much easier for me. I can focus on getting one dimension "right" (light vs dark) before I think about color at all.

I think this greyscale-then-apply-color workflow is one example of a workflow that is made practical with the help of digital tools. Although greyscale-then-apply-color was possible in traditional physical media, too (painters have long done monochrome underpaintings and glazed color over them), it was slow and very hard to undo or control.

Meanwhile, when we use digital tools, color can go on as a separate layer and be swapped out pretty quickly.

Perhaps the traditional painters of the ancient centuries, upon seeing us indulge ourselves in Photoshop and the evils of its layers feature and Ctrl+Z, might balk at our use of technology to cover for our inability to do (2) and (3) together. Perhaps the art student who dares to slap a color layer over a greyscale painting is an artist who is ""deficient in the rare, trained eye that sees value and hue together as one.""

And yet you hear very few artists lamenting,

"Alas, the youth of these days no longer know how to mix a convincing shadow tone straight on the palette without the hand-holding of that wicked layers feature and Ctrl+Z, and what will become of our artists if this sacred instinct is bred out of the population.............?"

...because digital art tools have sort of normalized the idea of splitting the work into separate passes, where artists who are less skilled at handling light and color in one pass can now break it into steps they can do well.

Could Gesture-First Be the Next Greyscale-First?

I think a similar argument could perhaps be applied to AI art.

If we think of the layers feature in drawing software as enabling greyscale-first art, by breaking rendering into two steps of value and then color, perhaps we can think of generative AI as a tool that enables gesture-first art, by breaking the drawing itself into two steps of deciding pose and expression, and then rendering it all into a finished image.

In the greyscale-first artist's case, they know they struggle with handling light and color in one step, so they break it down into the steps of (1) define the value by rendering in greyscale, (2) define the color by applying a palette on a separate layer, both of which are steps that they are individually good at.

In my gesture-first art case, I know I struggle with drawing a finished panel in one step, so I break it down into the steps of (1) define the pose and expression with gesture drawings and lineart, (2) render in full detail with AI.

My strength is in storytelling and in drawing poses and facial expressions, and that's where I'd claim my creative input is, not in the rendering. I'm less sad about offloading the latter to AI (meanwhile, I would be appalled at creative works where the characters were fully posed and written by AI).

Of course, the layers feature and its democratization of greyscale-first art didn't mean that everybody started producing beautiful art overnight, either because

• they weren't aware of this greyscale-first technique in spite of having easy access to the layers feature, OR
• they didn't have enough practice doing it, so the transition from value to color still looked clumsy, e.g. muddy shadows or physically impossible lighting

I think the same goes for AI art. It isn't inherently bad; it only looks bad when it's used clumsily. For example, AI art looks bad and super uncanny and soulless when its user doesn't know to decide pose and expression first, and also fails to correct the parts that are physically impossible, but can still look human and soulful when its user is an artist who has drawn the poses and expressions themselves.

So when you use AI in more thoughtful ways, it can actually enhance your work. The more you supply the AI with your own human-produced drawings, the more soul your work will have, and the more it will stand out from the work of someone who blindly prompts.

"Okay, but what about the artists whose work trained these models without their consent? What about someone who rips an artist's drawing, feeds it to generative AI, and calls the result their own? What about the illustrators losing commissions and jobs? Did you ever think about that?"

Yes, yes, and yes! These are bigger questions than the "is gesture-first still your own work?" one I spent my whole post on.

They deserve more than a hasty paragraph at the bottom of a "fun" post, so I'll save them for a later post.