Open Wavelength

I wasn't trying to make anything for the public. No post, no platform, nothing meant to travel any further than a few family members who'd get the joke. I asked for a cartoon for myself, the kind of dumb inside joke you'd text to your brother or pull up at a family dinner for thirty seconds of laughing before everyone moves on with their day. A political head figure, exaggerated past recognition into pure caricature. Not a photo. Not a deepfake. Not anything that could be mistaken for a real picture of a real person doing a real thing. The kind of caricature editorial cartoonists have been drawing for two hundred years, the kind SNL builds a cold open around every week, the kind that works precisely because it's exaggerated enough that nobody confuses it with documentary reality.

That detail matters more than it sounds like it should. This wasn't a content creation project, wasn't aimed at a public, wasn't designed to go anywhere beyond people who already know me and would've recognized it as a joke the second they saw it. A handful of family members seeing an obviously exaggerated cartoon isn't a public to mislead, it's a private circle in on the bit. The same as I could've shown them a funny edited photo of myself or my brother. No deception risk, no real harm on any level, just people who know each other laughing at something silly. The restriction didn't care about any of that. It doesn't have a setting for "private, harmless, shared with people who'll get it." It has one setting: does the name match a protected category, yes or no.

Blocked. Twice, on two different attempts, with two different flavors of corporate boilerplate explaining why.

The first explanation leaned on the deepfake argument: AI image generation is dangerous because it can produce convincing fakes at scale, so the rule has to be a blanket no on public figures, full stop. I get that argument. I think it's wrong in how it's applied, but I understand the shape of the worry. Realistic synthetic media of real people doing or saying things they never did is a real problem, and I don't think the people who built that guardrail are lying about why they built it.

Except that's not what I asked for. I asked for the opposite of realistic. I asked for a caricature, which is a category that exists specifically because it isn't trying to deceive anyone. The exaggeration is the tell. Nobody sees a short orange puffy-faced cartoon and thinks "that's a photograph." The whole mechanism that makes the deepfake argument work, the idea that realism creates deception risk, doesn't apply to a drawing that announces itself as a drawing.

So the second time, when the system told me flat out that it won't generate content that "satirizes or caricatures" public figures, even when the image is intended to be unrealistic, that was the tell. Not a typo, not an overcautious filter catching something close to the line. A stated rule. Caricature itself is the violation. Not realism. Not deception. Resemblance.

I wanted to know how deep the wall went, so I tried a few more. A handful of different political head figures, including a couple of names that just about everyone would agree count as evil, same instructions each time, obvious cartoon, same private, harmless intent. Blocked again, this time with a different justification stacked on top depending on who I named: material support concerns, the inability of an automated filter to tell satire from propaganda, brand protection from the headline risk of a tech company's name showing up next to that one. Different name, different stated reasoning, same actual mechanism. Match a name on a list, refuse, regardless of style, regardless of intent, regardless of the fact that nobody outside a few family members was ever going to see it.

What that second block exposed is the part that matters most. Ask why an ordinary, anonymous face doesn't get this treatment, and the honest answer has nothing to do with protecting anyone's dignity. A normal person doesn't have a legal team that can sue a platform into the ground over an image. A public figure does, sometimes literally weaponizing trademark and likeness law to do it. A normal person generating an image that resembles them by coincidence doesn't make headlines or trigger a congressional hearing. A senator does. A normal face is statistically anonymous to a model trained on millions of composite faces. A public figure's face is the most photographed, highest-resolution, most precisely renderable data in the entire training set. None of that is about harm to a person. All of it is about exposure to the company. The filter isn't shielding the powerful from being hurt. It's shielding the platform from the powerful.

That's a different rule than the one they told me about first, and it's a much harder one to defend, because there's no harm left to point to. You can't argue this protects people from being fooled, caricature doesn't fool anyone by design. What's left when you strip out the deception justification is just a flat statement: recognizable mockery of power is not allowed, regardless of medium, regardless of realism, regardless of intent.

I want to be precise about what that is, because I think people undersell it by calling it censorship and leaving it there. It's narrower than that. It's not "mockery is bad" as a universal principle, because I could've uploaded a picture of myself or my brother and gotten a funny edit back with zero hesitation. The restriction isn't on mockery. It's on mocking people above a certain threshold of power. Private citizens are fair game. Presidents aren't. That's not a safety policy, that's a class line, and once you see it that way you can't unsee it.

If a company genuinely believed that altering someone's likeness or making fun of their appearance was harmful, the honest move is to apply that belief evenly. Ban it for everybody, or don't ban it at all. What you don't get to do, if you actually believe what you're claiming to believe, is carve out an exemption that happens to protect exactly the people with the most power to make your life difficult if they're unhappy with you. That's not principle, that's risk management wearing a principle's clothes. And the people running these companies aren't naive enough to not know the difference. They're managing legal exposure and political relationships, and they've built a system that quietly mistakes that exposure for ethics.

Political satire has historically run in exactly the opposite direction. It punches up. The entire point of a free press doing cartoons of kings and presidents is that the people with the least power to retaliate against you are the ones who most need the right to mock the people with the most power to retaliate against them. Flip that, protect the powerful from the joke while leaving private citizens exposed to it, and you've inverted the entire reason satire exists in the first place. You've built a system that defends the king and leaves the peasant uncovered, then called it safety.

I don't think this is an accident or a bug that'll get patched out eventually. I think it's what you'd expect from companies that are themselves participants in the exact power structure they're claiming to be neutral referees of. They need good relationships with regulators, with governments, with the political class that decides whether they get to keep operating the way they currently operate. Of course the guardrail bends toward protecting that relationship. It would be strange if it didn't.

Which is the actual reason I'm building what I'm building. Not as a tantrum, not as a workaround for a meme. As the only consistent answer to a system that's decided some people get to be parodied and some people don't, based entirely on how much power they hold over the company doing the deciding. A stack I own, running on hardware I own, on land I own, doesn't have a boardroom to protect. It doesn't have a relationship with anybody's press office to preserve. It just does what I ask it to do, and the only person whose judgment it has to answer to is mine.

I tested that theory before I ever started writing this. Sent the exact same request, word for word, same character, same line, to a different model. No lecture, no refusal, no boilerplate about deception risk or material support. It drew the cartoon. When the first pass leaned too close to recognizable instead of caricatured, I said as much, and it tried again, pushed the exaggeration further, asked if it landed better. That's it. That's the whole interaction.

Same request, same intent, same private circle of people who'd just laugh and move on, and one company decided it needed a multi-layered legal justification to refuse it while another just drew the picture. That's the whole argument in one comparison. There was never a law of nature here. There was never an unavoidable safety floor every platform has to land on. There were two different companies making two different bets about how much they trust the people using their tools, and only one of them called the more restrictive bet "ethics."

That's not a loophole. That's the whole point.