In my theory the overunity can be generated from thermal fluctuations by lowering their dimensionality at some point. If we put a charged capacitor inside the hot gas, then the molecules of gas will hit the capacitor and they will induce voltage spikes into it. This noise can be measured as a common thermoelectric noise within electronic circuits and it indeed decreases with temperature.

The problem is, this noise cannot be rectified and utilized in any way, once the rectifying diode remains as hot, as the gas, the energy of which we are just trying to utilize. Once we will cool this diode, then yes, the noise of diode will be lower than the thermoelectric noise and we would get some voltage on it. The problem is, such an energy production will be just another form of thermoelectric pile, because the diode will gradually heat itself just with the electric noise, which it rectifies. So we must cool it for to keep the production of energy running and this cooling would also require an energy.

In this respect it's interesting, that the we can drain a quite substantial energy flux by cooling surface with cosmic space. Recently such a surface has been constructed by layering of nanosurfaces and it keeps cool itself between 4 and 5 degrees below the surrounding air temperature - which is already temperature difference worth of utilization with classical thermoelectric Peltier cell.

For example, Ann Makosinski did use power converter integrated circuit, LTC3108 for it. The circuit contained FET’s that would oscillate at voltages as low as 20mV. When used with a recommended   transformer, the IC would provide well over 2.5 volts AC. Ann got for 50 mV DC from the Peltiers about 6 mA at 5 Volts AC which was sufficient to light the LED. The result was a bright light at just 5 degree Celcius of Peltier differential.

But such a way of energy generation isn't still free energy in classical sense, overunity the less. We are just draining an energy from accelerated cooling of Earth with cosmic space, i.e. we are still utilizing solar and geothermal energy - just in indirect and diluted way not worth the effort.

In my opinion the situation with utilization of thermoelectric noise will change, if we wouldn't collect voltage fluctuations with 3D resistor, but with some thin narrow 2D plate (graphene layer) or even 1D nanowire (carbon nanotube). The trick here is, once we constrain electrons in their motion in one dimension, then the energy of their vibrations increases in remaining dimensions. The thermoelectric noise at the end of such nanowire will be therefore larger, than at previous 3D case and we can rectify it even with diode at the same temperature. Recently so-called time crystals (a miniature perpetuum mobiles in essence) were observed, which are working on the same principle (constraining motion of charged particles with Penning or magnetic trap).

Because the PN junction of diodes is actually quite narrow, even common LED diode can exhibit a tiny overunity, once it gets polarized in the proper direction. Of course, such an effect is quite weak at the case of single diode - but we can multiply it with usage of graphite nanolayers or nanotubes suspended within charged electret - which is IMO the principle, on which the Steorn Orbo Cube technology was working.