I explain my vote that it is possible:
Stefan-Boltzmann's law (1879) says that radiative power is given by the formula:
Pr = sigma.jpg × T4 (with T: absolute temperature in K)
with sigma = 5,670 400 × 10-8 Wm-2.K-4 (Stefan-Boltzmann constant)
So at 27 ° (300K), a black body of one square meter emits: 300 * 300 * 300 * 300 * 5,670400 × 10-8 or about 460 W
If the box is one m2 and is almost a blackbody with an emissivity of 0,9, it emits 414 W inwards.
Let's place a 0,2 emissivity photovoltaic sensor at peak wavelength at 300K which faces all surfaces of the box.
The sensor receives 414 W and emits 92 W. It generates 322 W and therefore its temperature increases. Not long since the radiation it emits increases to the power of 4.
If the sensor has an efficiency of 20%, it can generate 322 * 0.2 -> 65 W.
I said photovoltaic sensor: it is necessarily a sensor with a very small forbidden band as for lead oxides or germanium (0,5 V against 1,3 for silicon). Small band gap = possibility of capturing far infrared.
In other words, the voltage at the sensor terminal will be very low.
So if if if if if if if if if if if if ...
- we know how to build a box with an emissivity of 0,9 to 300K
- that we find a material which can convert far infrared photons (surely not pure graphene - but why not associated with dyes) which has a low emissivity at room temperature
- that we arrive at an improbable return of 20%
- we manage to produce 60 W per m2 maximum
Note: no need for a box, just place the sensors facing a concrete wall. This is the principle of the infrared camera which we know that we do not harvest much energy.
You can also do it by playing on the emissivities with thermocouples. it produces energy - but jokes.
So far we have only succeeded in capturing energy under these conditions by cooling the sensor with more energy than collected.
In a closed box no thermal machine can work. It is necessarily by using radiative emission and by playing on emissivities.