The All-Sky Automated Survey for Supernovae (ASAS-SN) is the first optical survey to monitor the entire sky, currently with a cadence of <~ 24 h down to g <~ 18.5 mag. ASAS-SN has routinely operated since 2013, collecting ~2000 to over 7500 epochs of V- and g-band observations per field to date. This work illustrates the first analysis of ASAS-SN's newer, deeper, and higher cadence g-band data. From an input source list of ~55 million isolated sources with g<18mag, we identified 1.5 * 10^6^ variable star candidates using a random forest (RF) classifier trained on features derived from Gaia, 2MASS, and AllWISE. Using ASAS-SN g-band light curves, and an updated RF classifier augmented with data from Citizen ASAS-SN, we classified the candidate variables into eight broad variability types. We present a catalogue of ~116000 new variable stars with high-classification probabilities, including ~111000 periodic variables and ~5000 irregular variables. We also recovered ~263000 known variable stars.