Doin' something different this time.
(A walkthrough of my AI-assisted comics process)
Gonna show you some of the progress on something before it's done, since I don't think many people actually read my under-the-fold process write-ups, and some might find this interesting.
I'm currently working on a second excerpt-story from the TyrannoMax & the Warriors of the Core comic series. In this case, it's the origin of the Cold Shoulder, and a side caper by GunBonnie & Clydillo robbing a bank and going up against PteroDarla and DeinoSteve.
Yes, GunBonnie and Clydillo were invented for a goof here on tumblr. They're characters now. This is how my process works. Might do a walkthrough on their design process at a later date.
To start, here's the first two pages in the story, as they are right now:
Page 1 is at the "flats" stage, while page 2 is just "inks." I tend to do a rough dialog layout first, shuffling and adapting as I go. This is a kind of one-man old-Marvel method (rough story -> pictures -> dialog built around pictures.), done in by-story-beat chunks.
I keep my dialog in a layers-folder, and apply a white stroke to the whole folder so I can see the dialog while working, and easily hide/reveal the whole thing.
"PENCILS" - Where the robot does the lifting.
A lot has changed since the first comic. For one, character consistency is now far easier. I've been using vidu for this one, which may sound odd as it's mainly a video generator, but they have a very strong character reference system, and it works with their image-generator.
This made the process a bit simpler, but it didn't correct it entirely. GunBonnie's arms and DeinoSteve's feet are still going to need frequent fixing, and I rarely get the exact pose I want, so my preferred 'colorforms' approach remains.
Basically, I take what I want for a given scene, and I generate using the character reference. Page 1 used parts of three gens:
Page 2 used these (among others):
I assemble those parts into photoshop, essentially doing the same thing one would do with any photobash/imagebash comic. Once I have them placed, I start -
"INKS" - Where things get normal.
For this, I take the layer for each color image, and I reduce that down to the linework (don't generate in B&W because it produces B&W comic inks, which are different than color comic inks). I do this by working 2x the normal 2x scale (total 4x) and using the threshold command. But because line work varies and colors get in the way, this means layering multiple passes at different intensities in different areas, then blending with manual inking. This is where I fix flaws, extend lines that don't go far enough, and do the same kind of fixes you'd make with sloppy pencils.
If the line quality is too thin, I tend to either select the lines, expand by x amount of pixels, and use the pencil to fill in selectively, or I duplicate the too-thin inks, offset them on a multiply layer, and erase areas where they get too thick. If an area should be pure shadow, I fill it in black.
Those using a stylus and whose hands work better will probably have an easier time with the manual parts of the process. I'm doing this with a mouse because that hurts the bones least.
Where I can, I keep these in layers so I can move characters around a bit if I need to for dialog. My panel gutters are a separate layer up-top which gives me some wiggle room around the edges.
This stage is basically normal digital inking with tight pencils, assuming you're penciling at all these days. (I can't use styluses, it both hurts and doesn't gel in my brain.) I guess most folks would just digitally do it straight to inks.
FLATS - The quotes are no longer needed.
At this point there's nothing different from the normal digital comics process. This is just my process for doing digital color:
Flats are just what they sound like, flat colors that serve as the base for final colors. Step 1 is shown here part-way-through, and it involves placeholder colors on a layer above a copy of the inks. If the inks aren't already shock black-white "thresholded", I make a copy of the inks and do that to it. Anti-aliasing is something I'm avoiding at this stage, as it makes things a lot more obnoxious if you need to make changes.
I use bright placeholder colors for white and near-white objects, since it makes it easier to spot gaps.
Because comic inks aren't (or shouldn't be) like a coloring book, the paint bucket tool is not going to cut it alone. Again, those with more functional hands will likely opt to just paint them down, I close off gaps with the pencil tool and then fill (which you can see some of in the screenshot above.)
The initial result of this (below, left) won't work on its own, it has the risk of creating white seams during down-scaling and layer collapse, and it means certain tricks in later stages will be harder to do. So I correct this by first selecting each color with 'contiguous' off and by expanding by 1-2 pixels depending on how thick the line widths are. I do this for every color in the composition, and move the gutter-placeholder to a separate layer above the flats.
I have a script I made that duplicates the layer called 'flats' four times, then nudges each layer 1 pixel, in each of the four cardinal direcations (up, right, down, left), then collapses those together and renames the result 'flats' so I can do it again. I use this to close up seams, filling in areas with the paint bucket as it becomes viable, until something like the above middle is done. (this is one of those things anti-aliasing makes a pain, and the previous step of manually expanding each color keeps them from making weird patterns at certain connection points.)
The work down on previous steps makes it a lot easier to take out my handy-dandy 1982 DC comics pantone guide and replace the colors with the ones I want to use in the final version.
Not taking it past flats today because the issue isn't to that point yet. I like to do full colors and lettering as a single block task since it makes it easier to keep consistency. So I'll hit with some how-to on the final coloration, lettering, and faux-print effects, possibly also the process of making period parody-ads, at a later time.
You can do this with any AI image generator, but those that can use references consistently (like Vidu) are going to save you some hassle. You can likely do this with a chat-based interface, but you'd have to start completely over with every image, as all the chat-based ones pollute future images with the totality of the previous discussion.
Hope this was interesting/helpful.
