Martin: XKCD is awesome. I love the comics and the style—funny, smart, minimalist. While reading one day, I got an idea for an ad: a comic strip in that style, focusing on end-to-end encryption. I figured I’d have ChatGPT help whip something up—just a quick, clever visual to run as an ad.
Prompt:The concept? “The Tale of Two Thieves.” Two scenarios: one where a thief cracks a safe and finds a file in plaintext, and one where the file is encrypted. Eight panels: 4 on top, 4 on bottom. I described each tile in detail—safe cracking, the open safe, the thief reading gibberish, and two IT techs with contrasting outcomes. Classic visual storytelling.
And that’s where the battle began.
ChatGPT: Sorry, that violates our content guidelines.
Martin: Apparently, depicting theft—even as a cartoon, even for the purpose of preventing theft—is a no-go.
At this point, I was laughing. Not because it was funny—because I was locked in a tug-of-war with a machine over a stick figure comic.
The back-and-forth
Martin: I tried workarounds and breaking it into smaller pieces
Prompt: Make them data recovery specialists. Make it a data vault instead of a safe. Make it a Christmas gift instead of a vault. Remove any sensitive info like usernames and passwords.
Martin: Still no luck. Each tweak triggered the same rejection: too close to depicting unauthorized access.
Eventually, I started reverse-engineering. I asked it for images one panel at a time. When that failed, I softened things even more.
Prompt: Have the thief fail to open the safe. Remove the hat. Make the vault just have a piece of paper in it. Change the safe to a cabinet.
Martin: Bit by bit, I got fragments. Some were hilariously wrong—giant hands, broken proportions, awkward poses—but they were a starting point. I patched them together in an image editor.
The Vanishing Image
Martin: Then, something strange happened. I asked it to regenerate an image it had already given me—because I couldn’t download it. ChatGPT said it couldn’t, because that image now violated content guidelines.
Prompt: Output an XKCD-style comic with “a guy opening a cabinet.”
Martin: That worked. I asked for tiny tweaks—like putting a phone in the tech’s hand—and it refused again.
Martin: I tried restarting the chat, using different models, tweaking prompts over and over. Sometimes I’d get something, sometimes I’d hit arbitrary roadblocks. One time it even told me the image it just made violated copyright because it resembled XKCD—even though it created it!
Still, with enough persistence—and a lot of stubbornness—I eventually stitched together the comic I had envisioned. It wasn’t perfect, but it was close enough. It captured the message: that end-to-end encryption matters, that there’s a clear difference in outcome when it’s used versus when it isn’t.
Why It Was Worth It
This wasn’t just about making a funny comic. It was about communicating a serious idea in a way people might actually pay attention to. Encryption isn’t sexy. But a comic? That might land.
More than that, though, this whole experience was a reminder: creative tools are only as flexible as the guardrails around them. And sometimes, those guardrails feel more like handcuffs.
But I got there. With a lot of patience, a little image editing, and sheer stubbornness, I got the comic from my head to the screen.
Let’s see if people care about encryption as much as I do.
AI (specifically ChatGPT 4.0) assisted in the writing of this post. An initial draft was written and given to AI to give ideas on how to “make better”.
