The benefit is that your audience knows whether it's a real picture of a thing that exists in the world. I wouldn't argue that's a particularly large benefit - but I don't see why labeling generated images would be a particularly large cost either.

I'm approximately a free market person. I hate regulation and believe it should only exist when there is a involuntary third party externality.

My position is that there in an unspecified benefit, the only cases specified here already are covered by other laws. All such generative labeling would do is increase costs (marginal or not, they make businesses less competitive) and open the door for further regulatory capture. Furthermore, refardless of commerciality, this is likely a 1A violation.

There are already laws against murder, but this doesn't stop communities from passing new laws when a cop gets murdered.

These arguments hold no water.

True. It just seems unnecessarily redundant.

I still forsee 1A issues. What about user uploaded content?

The map is not the territory. No photo represents a real thing that exists in the world. Photos just record some photons that arrived. Should publishers be required to disclose the frequency response curve off the CMOS sensor in the camera and the chromatic distortion specifications for the lens?