Another thought: use a three-frame gif to show a password to unlock a cashu qr token. Each of the three frames could show a part of the stroke of each letter. For example, frame 1 could have the left slant on an A, frame 2 could have the horizontal part of letter A, and the third frame has the right slant. The time between gif frames could be only 50ms so the eye couldn't even tell its animated, but a computervision couldn't take advantage of the eye's visual integration time.