What stops one from making a more secure pixelation algorithm?

One that will shuffle around or maybe just randomly derive the pixelated blocks if they're within a certain color threshold of similarity?

I still generally like pixelation over drawing a box for aesthetic purposes.

Why bother? Just have the tool put random text to fit then pixelize that.

Well you can increase the size of the blocks, maybe to full font-height. But at some point in future, some ML-guy will even decipher that probably.

For elegance, it's a bit artificial, being tied to the technology of things that use pixels and popular software that zooms in on pixelated images as a grid of single-color squares. Maybe some unintelligible character-like symbols would look more elegant?