To summarize the previous threads on this, self stability of a bicycle relies on a geometry that steers the front tire into the direction of a lean. The conventional method for doing this is trail, where the extended steering axis intercepts the pavement in front of the contact patch. The alternative used in some experiemental bikes locates a mass ahead of and above the front tire, with the front tire mounted so it's free to rotate about it's steering axis, resulting in a yawing torque on the frame when leaned, which ends up steering the front into the direction of lean, without requiring any trail or caster setup.
There is a small gyroscopic steering reaction due to the roll torque related to the lean angle of a bike (gravity effectively pulls down at the center of mass, pavement pushes up at the contact patches), but that reaction normally doesn't steer inwards enough to result in self stability (note the roll torque goes to zero when the bike is leaned but in a coordinated turn), so the main effect of gyroscopic reaction is to dampen the lean rate by opposing the corrective steering input due to steering geometry.
At high speed on a motorcycle, around 100 mph == 161kph or more, the gyroscopic dampening dominates and mathematically causes a bike to fall inwards at an extremely slow rate, called capsize speed, and the perception to the rider is that a motorcycle at high speed tends to hold a lean angle unless / until the rider uses
countersteering to correct or change the lean angle.
On a side note, a skilled rider can balance a non-moving bicycle with trail using the conventional steering inputs, but the movement of the contact patch with respect to the frame of the bike is limited, so only small lean angles can be corrected. A rider can also shift weight and/or hop to maintain balance on a bike, and this is done in "trials" type competitions to reorient (rotate) a bike while on top of a small surface.