check out StegoAppDB.
This is a HUUUUGE database of stego & original cover images used to train algorithms to identify steganography.
There are many machine learning stego projects up on github. Clone one of those bitches, import your clean set and your steg set for training… then go grab a bunch of "suspect images" from whatever board you want, and run them against your newly trained algo.
The one I suggest trying is PixelKnot. PixelKnot is an Android app that uses the F5 algorithm to do its thing. There is well over 300gb of PixelKnot pictures to play with on StegoAppDB, free to use & download.
You can obviously create your own sets as well by scripting tools, but this is an easy way to get started with KNOWN images.
#HappyHunting