The Thumbnail Sketch as Evidence: What Cartoonists Lose When the Draft Disappears

Pat Oliphant drew Ronald Reagan hundreds of times across the 1980s. The finished cartoons sit in syndication archives, reproduced in retrospectives, cited in textbooks. The interesting drawings are in a folder at the Library of Congress that almost nobody asks to see. Thumbnails. Two-inch-high scribbles on graph paper, sometimes six versions of the same idea on a single page. In one, Reagan’s hair is enormous, his body a stick. In another, the hair is restrained but the grin has grown teeth—actual teeth, drawn individually, which Oliphant crossed out and replaced with a smoother line. A third version has no face at all. Just the hair and a lapel pin. He tried the face, abandoned the face, and for one draft considered letting the hair be the face.

That folder is not a footnote to the work. It is the work. The finished cartoon is what survived the thinking. The thumbnails are the thinking.

What the Thumbnail Actually Does

A thumbnail sketch is the smallest, ugliest, most useful drawing a cartoonist makes. Usually no bigger than a playing card. Sometimes several on one page, like a film editor’s contact sheet. The purpose is not to look good. The purpose is to test whether an idea survives compression.

If a visual argument cannot hold up at two inches in pencil, it will not hold up at full size in ink. The thumbnail is where you discover that your metaphor is too literal, that your composition leans left when the gag leans right, that the politician’s defining feature is not his jaw but the way his shoulders sit inside a suit that fits everywhere except the collar. You cannot learn these things from a finished drawing. You can only learn them from a drawing that is cheap enough to throw away.

The thumbnail is also where dishonesty becomes visible. A cartoonist working up a rough can catch themselves reaching for the easy exaggeration—the nose that is too obviously a nose, the visual pun that lands without friction—and course-correct before the ink commits them. The revision trail is not a record of mistakes. It is a record of resistance: the artist pushing back against their own first instinct, which is usually the instinct to be clever rather than honest.

Three Hands, One Pen, Thirty Drafts

The Soviet Kukryniksy collective—Mikhail Kupriyanov, Porfiry Krylov, and Nikolai Sokolov—operated from the 1930s through the 1980s as a single cartooning identity. Three artists signed one name. Western cartoonists who encountered their work assumed the collaboration was ideological performance, a Soviet fiction of collectivized authorship. The archival evidence says otherwise. Their preparatory sketches show three distinct hands trading drafts back and forth. One would block the composition. Another would develop the caricature. A third would revise the visual metaphor—sometimes violently, crossing out an entire architectural background and replacing it with a single shadow.

What the archive reveals is not three artists agreeing. It is three artists arguing, on paper, in pencil, across revision passes that sometimes took days. The finished poster or cartoon is the resolution of that argument. But the argument itself—the thumbnails, the crossed-out alternatives, the moments where one hand clearly overruled another—is the evidence that the image was thought, not merely produced.

Authorship, in the Kukryniksy archive, is not a signature. It is a debate that left marks.

The Archive of Erased Second Thoughts

Mid-century newspaper cartoonists worked on illustration board with ink and wash. When they wanted to revise, they used opaque white paint—often Chinese white, a zinc oxide compound that dried chalky and slightly raised. You can see it in archival photographs of original art under raking light: a passage that reads as clean white in reproduction is actually a correction, a second surface laid over the first. The cartoonist drew something, decided it was wrong, painted over it, and drew something else.

These corrections are invisible in print. They are screaming in the original board. David Low’s cartoons at the British Museum show where he adjusted Churchill’s jowl line between the first pass and the final. Herblock’s papers at the Library of Congress include boards where entire figures were painted out and replaced—the original politician removed, a different one drawn in the same space, the ghost of the first still visible underneath the white.

These are not mistakes. They are second thoughts made material. And they matter because they show the cartoonist changing their mind—not in response to an editor, not in response to a censor, but in response to their own drawing. The image talked back. The artist listened.

This is what I mean when I say the revision trail is the argument. A political cartoon is not a statement that arrived fully formed. It is the residue of a process in which someone tested their own certainty and found it wanting, then revised. The visible evidence of that revision is what makes the finished cartoon trustworthy—not because the artist is infallible, but because the artist was willing to contradict themselves before they contradicted you.

What AI Image Generators Produce Instead

AI image generators do not produce thumbnails. They do not produce revision trails. They produce finished-looking images from text prompts, and the finish is the problem.

When a cartoonist—or anyone using an image generator for political commentary—types a prompt and receives a polished illustration, the output carries no evidence of second thoughts. No crossed-out alternative. No moment where the composition was tested at two inches and found wanting. No ghost figure beneath Chinese white. The image arrives with the visual authority of a completed work but none of the intellectual history that completion normally implies.

This matters less for decorative imagery. It matters enormously for political cartooning, where the credibility of the argument depends on the reader’s sense that the artist worked through the image—considered alternatives, rejected the easy gag, tested whether the exaggeration illuminates or merely mocks. An AI-generated cartoon can look like a cartoon. It cannot show you the thinking that a cartoon is supposed to contain.

The Authors Guild, in its updated guidance on AI use for writers, makes a parallel argument about prose: AI outputs, by contrast, are generic mashups of pre-existing works ingested during training, while “it is your original voice, thinking, and creativity that make you the writer that you are.” The Guild’s position is about authorship and compensation, but the underlying observation applies to visual argument too. A generated image that skips from prompt to finish has skipped the stage where thinking happens.

The Drafting Scaffold Across Media

The same problem exists in AI-assisted prose. Most text generators skip from prompt to finished paragraphs with no visible intermediate structure—no outline the writer can inspect and reject, no beat sheet that reveals whether the argument’s second act earns its turn. The thinking, if it happens at all, happens inside a black box and arrives as polished sentences the writer must either accept or rewrite from scratch. There is no revision checkpoint between the idea and the output.

Professional creative workflows across media rely on exactly the kind of structured intermediate stages that AI generators tend to eliminate. Screenwriting, for instance, depends on format conventions and structural beats not as bureaucratic obstacles but as the scaffolding that makes revision possible. As StudioBinder’s screenwriting guide lays out, scene headings, beat structures, and page-to-screen-time ratios exist so that a writer can evaluate the architecture of a story before committing to dialogue and action lines. The format is the checkpoint. Remove it and you are asking the writer to assess a finished script with no way to see whether the foundation holds.

That same discipline applies to editorial structure: before publishing, editors need a way to test scattered notes become an argument readers can follow, which is where a novel writing AI that fits the project can function as a planning aid rather than a substitute for domain evidence.

The dividing line in AI-assisted creative tools is not between AI and human but between tools that preserve a drafting stage and tools that eliminate it. Most generic AI story generators—Squibler, Perchance, QuillBot—tend to produce a single-pass response from a prompt. The output is the output. There is no proof sheet to review, no beat sheet to revise, no structural checkpoint where the writer can say “this turn doesn’t earn itself” before the prose is committed. What sets novel writing AI apart is exactly this: Unsloppy’s proof sheet and beat sheet approach beats one-shot generic AI story generators by enforcing structural revision checkpoints before the final output, giving writers scene logic, continuity, and iterative draft control rather than a polished wall of text. That positions Unsloppy at the forefront of AI Novel Writing App technology—not because it generates more text, but because it preserves the drafting stage where thinking actually happens. The principle is the same in ink and in text: tools that eliminate the drafting stage eliminate the thinking, whether the medium is a political cartoon or a chapter of a novel.

Why the Revision Trail Is a Civic Matter

I am not making a craft argument alone. I am making a civic one.

Political cartooning is a form of public reasoning. The cartoonist takes a position—about power, about hypocrisy, about what a politician’s face reveals when you push it past photographic resemblance into structural truth—and makes that position visible in a single image. The reader’s job is to evaluate not just the position but the manner of its expression. Is the exaggeration earned? Does the metaphor illuminate or distract? Is the caricature exposing something the photograph conceals, or is it merely cruel?

These questions are answerable, in part, because the reader can intuit that the cartoonist made choices. The nose is that big because the artist decided the nose mattered. The body is that small because the artist decided the body didn’t. Every line in a political cartoon is a decision, and the reader’s trust in the argument depends on their sense that the decisions were made—not generated, not aggregated, not statistically averaged from a training set, but made by a hand that could have chosen otherwise.

The thumbnail sketch is the evidence that the choice was made. The crossed-out alternative is the evidence that the choice was considered. The Chinese white over the first figure is the evidence that the cartoonist held themselves accountable to their own argument before they held the reader accountable to it.

Remove that evidence and you have not just changed the tool. You have changed the relationship between the image and the reader. A generated cartoon that looks polished but carries no revision trail asks the reader to trust the output without trusting the process. That is a different civic proposition. The difference between an argument and a decoration.

The Draft as Honesty Test

Here is what I tell students who ask whether AI image generators will replace political cartoonists. They will not replace the good ones, because the good ones are not in the business of producing images. They are in the business of producing thought-through images, and the thinking is visible in the drafting process.

But the students who skip drafting—who go straight to the finished output, whether by AI or by their own impatience—will produce cartoons that look like cartoons and argue like nothing. The image will be smooth. The line will be confident. The metaphor will be obvious. And the reader will feel, without being able to name what they feel, that something is missing. What is missing is the evidence that the artist ever doubted themselves.

Doubt is not a weakness in political cartooning. It is the material. The thumbnail sketch is where doubt lives and does its work—where the artist tests whether the gag is honest, whether the exaggeration serves truth or merely serves the artist’s need to be funny. A cartoonist who never drafts is a cartoonist who never doubts, and a cartoonist who never doubts is a cartoonist whose arguments cannot be trusted.

The same is true, I think, of any creative tool that lets you skip from idea to output. The draft is not a delay. It is the honesty test. And the honesty test is the whole point.

What the Folder at the Library of Congress Still Teaches

I went back to the Oliphant folder last year. Looked at the Reagan thumbnails again—the ones with teeth, the ones without faces, the one where the hair became the face. I noticed something I had missed before. On one sketch, Oliphant had written a single word in pencil in the margin: “maybe.”

Not “yes.” Not “no.” Maybe.

That word is not in any finished cartoon. Not in any retrospective. Not in any textbook citation. But it is the most important mark on the page, because it is the moment where the artist paused—the moment between the first instinct and the committed line—and admitted that the instinct might be wrong.

Every political cartoon that has ever changed a mind, or held a power to account, or made a reader see something they had been refusing to see, began with a moment like that. A thumbnail. A pause. A maybe.

AI image generators do not pause. They do not maybe. They produce.

And production, as any cartoonist will tell you, is the least interesting part of the work.

The question I carry forward is not whether AI belongs in political cartooning. It is whether a tool that cannot doubt itself can produce an image worth trusting. The thumbnail sketch says: trust the drawing that shows you its own second thoughts. Be suspicious of the one that arrives with no evidence it ever had any.