The one problem with this approach is that it gives one guy with a big GPU (or even an ASIC) a huge advantage over "casual" users (e.g. mobile web). You could perhaps make the gap smaller by using one of the "ASIC-resistant" mining algorithms?

By the way, for making more efficient mining algorithms, it would be nicer if the nonce is on the end of the message, aligned to a sha256 block. As is, it encourages someone to set the real nonce to a fixed value, and append their own garbage (for the text message version at least, I suppose for the other formats it doesn't matter).

Edit: it would appear that your current mining algorithm is pretty inefficient. I seemingly managed to out-hash everyone with just 32 leading zeroes, taking a few Python-CPU-minutes to find. To cement my place further, I spent a few GPU-minutes to find 40 leading zeroes.

The current "best hash in grid" has 28 leading zeroes. If we pessimistically take that as the average, and there are 2^10 pixels, it would take 2^38 work to flood the whole grid right now. If I spent my GPU-minutes on grid pixels rather than the text, I could have flooded the grid 4 times over.