is the Hubble Space Telescope archive. Type in a target name (M57 is a pretty one; planetary nebulae work very well, the big Messier galaxies and globular clusters are a bit too big for the Hubble field of view), click on 'footprints' to get a picture of the sky with little boxes showing where the photo was taken, click on a footprint and go to 'images' to get a set of little pictures with the ones under the footprint highlighted in green, click on 'view online' to get a Google Maps-type zoomable view of the image.
This is more science data than it is pretty pictures, though it's not the rawest level - most of the images have been made from three exposures by a process that takes out the cosmic-ray hits, and (as you'll see in M57) some script has noticed that images have been taken of the same field in several filters and made a colour-combined version of that.
Selecting something like Cygnus A and seeing that the best pictures from the best instrument ever orbited still look like a slightly dirty orange blob, or looking at an ACS picture of a star-cluster in M33 (870,000 parsecs away) which looks rather like the photo I took of the Perseus double cluster (2300 parsecs away) with a 30mm wide-field lens last Saturday gives me a much stronger feeling that the universe is a very big place of which we know little than usual awe-driven articles about Hubble provide. The image-browser also makes it very clear that WFPC is a 2.4-megapixel camera; WFC3, being launched in about a fortnight, will be 16 megapixels and produce rather less noisy pictures.
I know HST time is insanely competed for, but I am a little surprised that people are so confident that galaxies are radially symmetric that they'll publish a paper on the distribution of star clusters in M33 having taken pictures of a dozen random non-overlapping fields covering maybe 15% of the galaxy. There's clearly sensible science behind 'we will take pictures of this elliptical galaxy in three UV bands to see if anything unexpectedly energetic is happening', but the UV detector is rather noisy and the galaxy doesn't emit much in the UV, so NGC4627 (picked as a random small galaxy from Saturday's APoD) is rather disappointing.