I found myself curious just how bad ASCII text would wind up after a run through a lossy compression scheme.
So I encoded 16,875 bytes of ASCII text into a .png image, converted it to a .jpg, then read the data back into a text file. See if you can decrypt the mess that resulted.
The SHA1 for the original file is 1B82F08A3AFC15DCDF808DC7C1F8A 309AD4EA5F7.
For the record, the jpg was made by GIMP, exporting at 100% quality, meaning that the mess here was caused by just the artifacts of changing a .png to a .jpg and back. The ASCII text after actual lossy compression was totally illegible even for very high compression qualities.