Zerochan.net is one of famous anime/game/CG imageboards with strong community and modest crossposting with other imageboards.
It has specific tagging system - close to e-shuushuu-net - but not to mainstream danbooru / safebooru / yande-re / konachan / sankaku.
That’s why Zerochan is a good distinct source for investigation of non-photographic images and their metadata.

This release devoted to paleonthologic part of board from very start till 31.12.2014 (ID=1820227) right before 2015-2016 release
The enormous volume of initial images (~650 GB) brought within reasonable limit with selective sampling described below.

Release contains:

• 313.044 JPG images in 183 zipped folders partitioned by 10.000-th ID
  • filtered by initial size
    ~ least(image_height,image_width)>=1080 – fullHD wallpapers as minimum
    ~ image_height*image_width>=1200000 – 1100x1100 included
    ~ image_width/image_height between 0.32 and 3.2 – not too disproportional
  • renamed “zerochan - id - up_to_3_sources ~ up_to_5_characters (up_to_2_artists).ext”
    ~ tags concatenated via “+”, spaces replaced with underscores
    ~ maximum file name length 220 symbols, characters tags may be truncated if too long
  • some gentle deduplication made
• metadata for every image ZERO_POSTS_2014.TSV in root folder 313.036 rows
  • from imageboard (original file URL, upload date)
  • image info (size, volume, md5 etc) both for original and sample
• tag info for Copyright / Characters / Artists and more ZERO_TAGS_2014.TSV with 1.815.947 rows
  • as parsed from site with Unicode suppressed / replaced
  • many of them used in file naming but there are a lot of more

About sampling:

Huge total size of initial images leads to unpractical torrent release - too big and not too worthy to be supportible.
I desided to selectively shrink big / bloated images to practical size with good quality, that was chosen as
1920 px longer side (Full HD both landscape and portrait) and JPEG quality 92%
I used image magick mogrify -thumbnail 1920x1920^> -quality 92 -format jpg and then
compare initial and sampled image to left initial image when negative or minor effect of resampling.
There are 90.375 original images left in release, as pointed out in IMG_TYPE column of ZERO_POSTS.

THERE ARE some - also sampled - rips on Sukebei tracker for Konachan and Yande-re. With nipples.