So basically a guy made a program to upscale anime images using "Deep Convolutional Neural Networks"

Sounds complicated? It is but i set up a demo site so you guys can test it out: waifu2x.booru.pics

More information on this project at: github.com/nagadomi/waifu2x

NEWS:
Link on images pages, left bottom to insta-waifu2x an image.

Just tested it.

Test 1: Made a test image consisting of a black border and a red circle. Tried to upscale it 2x with no noise reduction. Program failed to noticeably do anything beyond simply making each pixel 2x bigger. 1.6x wasn't much better, blurry pixelation instead of just pixelation.

Test 2: Used post #532721. Same settings as before. More successful. Areas with high contrast (eg. some parts of the collar) tend to come out blocky, antialised lines fared better.

Program creates jpeg artifacts regardless of what format the original image was. Always creates a png despite this, so expect bloated filesizes (Test 2 went from 431 KB to over 4 MB).

Conclusion: It works, but needs some improvement. It did an adequate (but slightly jagged) job with gradiants and antialiased lines, and made some faded outlines stand out more which was nice, but had difficulty with extreme contrast.

See if waifu2x.udp.jp/ does it any better, i haven't trained mine at all.

Test 1: Same result.
Test 2: Same result.

Side note: Why 1.6x instead of 1.5x?

Neat!

So on every picture on the site there is a link on the bottom left where you can click and it will auto waifu2x the image if its within the max size that we allow. This whole thing is in beta with a better interface then the original.

Hmm..I suppose that does solve a few problems.