Although fluids are pervasive among tectonic faults, thermo-hydro-mechanical couplings during earthquake slip remain unclear. In particular, fluid induced seismicity is today a subject of major interest for the geo-energy community. We report full dynamic records of stick-slip events, performed on saw cut Westerly Granite samples loaded under triaxial conditions (stresses σ1> σ2= σ3) at pressures representative of the upper continental crust (σ3~ 45-95 MPa) Three fluid pressure conditions were tested, dry, low, and high pressure (i.e. Pf=0, 1, and 25 MPa). Friction (μ) evolution recorded at 10 MHz sampling frequency showed that, for a single event, μ initially increased from its static pre-stress level, μ0 to a peak value μ p it then abruptly dropped to a minimum dynamic value μd before recovering to its residual value μr, where the fault reloaded elastically. Under dry and low fluid pressure conditions, dynamic friction (μd) was extremely low (~0.3 and ~0.2 respectively) and co-seismic slip (δ) was large (~250 and 200 μm respectively) due to Flash Heating (FH) and melting of asperities as supported by our microstructural analysis. Conversely, at high pressure (pf=25 MPa), μd was higher (~0.45), δ was smaller (~80 μm), and frictional melting was not observed on post-mortem surfaces. We calculated flash temperatures at asperity contacts including heat buffering by on–fault fluid. Considering the isobaric evolution of water’s thermodynamic properties with rising temperature showed that pressurized water controlled fault heating and weakening, through sharp variations of specific heat (cpw) and density (ρw) at water’s phase transitions. Injecting the computed flash temperatures into a slip-on-a-plane model for Thermal Pressurization (TP) showed that: (i) if pf was low enough so that frictional heating induced liquid/vapour phase transition, FH operated, allowing very low dynamic friction during earthquakes. (ii) Conversely, if pf was high enough that shear heating induced a sharp phase transition directly from liquid to supercritical state, an extraordinary rise in water’s specific heat acted as a major energy sink inhibiting FH and limiting TP, allowing higher dynamic fault strengths. Further extrapolation of this simplified model to geothermal reservoir depths shows that TP is the dominant weakening mechanism up to ~5 km depth. Increasing depth allows somewhat larger shear stress and reduced cpw rise, and so, substantial shear heating at low slip rates, favouring FH for fault weakening. Our results demonstrate that thermophysical properties of water should be taken into account in the formulation of the physically based model of earthquake forecasting.