Full threadreneberlin·compare it to jp2a / ascii-image-converter In ascii they perform very well without ML, of course. You could set up a training based on that.View on HN