Once I attempted to create a very efficient video compression tool without caring how lossy it was.
I didn't use any traditional picture compression technique such as DCT or movement detection but instead went for a very naive approach.
In short, every pixel is represented by one bit, defining if the colour is moving "closer" or "further" from a given colour taken from a predefined sequence that the codec knows.
Only a few pixels per "zone" are sampled on each picture and the decoded colours are blurred to prevent long-term artefacts to appear.
The final sequence of bits is then manipulated to make it easier to apply RLE compression on them.
Some versions of this algorithm were able to compress a typical full length movie into less than 50Mb, but many artefacts were appearing over time.
This video presents a compressed movie being uncompressed back to .avi with various versions and parameters of the algorithm
And here is the last version of it (processing sketch, not an "actual" codec).