The studio

Make a sealed code.

Write the question once. We encrypt it into the code, and you get something you can print.

Start from a recipe

Shown to whoever scans it, so they know the code is about the right thing.

Name the object, give the facts an AI cannot see, and ask for a shape: three steps, one paragraph, a short list. Unbounded questions get unbounded answers and nobody reads those standing up.
A sealed code carries the whole prompt, so length is module count: under about 250 characters scans from across a room, and much past that wants a tracked code.

With a key, the code is useless to anyone who does not have it. The decryption key is derived from what you type, so there is no check to skip. Write it down. If you lose it the prompt is gone, including for us. That is the guarantee working, not a bug.

Without this, a stranger who scans a locked code hits a dead end and assumes the code is broken. With it they get somewhere real that you chose, so the code reads as working. We never send them automatically and we always show them the address first, because a page that bounces people onward unasked is how phishing works. The address is encrypted in the code like everything else.

Sealed carries the whole encrypted prompt inside the QR itself. We store nothing: there is no row in a database with your prompt in it, and there is nothing for us to lose or leak. The trade is that the code is fixed once it is printed, and a long prompt makes a dense code. Tracked puts six characters in the QR and keeps the prompt here, which buys a small code, a much longer prompt, scan counts, and the ability to change what the code says after it is on the wall.

The single biggest lever on how many people scan it. Say what they get.

Your initials in the middle. Switches the code to the strongest error correction so it still scans with the middle covered.

Free. No account. You can scan the result with your own phone on the next screen.