>> Do Prime members still get unlimited photo storage?
> Yes.
Heh. Anyone tried writing an encoder to make files look like photos, and upload them to any of these "unlimited photo" services. I am sure they are probably watching for it and will close the account.
Then what if it there is a way to gradually mix in data with actual photos, a bit like in steganography. But more aggressive. And create what still looks photos, just like really noisy.
Dropbox did a thing a few years back where they gave you permanent additional storage for uploading photos. A jpeg header followed by gigabytes of zeroes supposedly worked just fine.
This is very easy to catch, though. Abusing comment fields might be more interesting. Use a small valid image, and insert 4MB of data into a comment field. It would be a fun project to implement this as a FUSE file system. I'm not sure whether any of the popular formats support arbitrary-length comments, and whether or not Amazon strips such data (it's image storage, after all, not file storage).
I wouldn't use steganography (it's not normally very space-efficient), just dump the payload data into a PNG's comment field. You can do this with exiftool "-Comment<=/path/to/inputfile" dummy.png and re-extract it with exiftool -b -Comment dummy.png > recovered_file. Of course you'd use libpng directly in a FUSE filesystem, but this shows that it's easy enough. Use a 20kB cat picture and 2 MB payload and you achieve 99% space efficiency.
It's probably inconvenient enough for most data hoarders for Amazon not to care. You can still switch to encoding it in the pixel values in case they do. I'm not familiar enough with PNG to comment on the space-efficiency of that.
Heh, back when Gmail was new and outrageous, someone wrote a GmailFS script or plugin or something, letting you store your files in a Gmail account. I guess it was one of the first cloud storage systems.
I wonder if it would be possible with machine learning to find out how data can be recovered after re-encoding an image using different encoders. This would ensure no loss of data. Least significant bit stenography must be very fragile.
To link someone else's project: https://github.com/vgel/flickr-fuse "Flickr-fs is a fuse filesystem that allows you to upload your files to flickr as encoded .pngs"
Although it does come with a fair warning that ...
"It's also really slow and liable to explode at any moment, so don't seriously use it."
not sure if anyone ever formalized it in some sort of library, but yes, it's been done! people were using flickr for storage years ago.[1] I have no idea if it's still possible.
Heh. Anyone tried writing an encoder to make files look like photos, and upload them to any of these "unlimited photo" services. I am sure they are probably watching for it and will close the account.
Then what if it there is a way to gradually mix in data with actual photos, a bit like in steganography. But more aggressive. And create what still looks photos, just like really noisy.