I think it would come down to the sourcing of the supposed evidence, if someone commits a crime in public they may well be recorded by multiple cameras from multiple angles, even if one camera purportedly showed a crime it could be proven to be false by a lack of corroboration from other cameras in the area recording at the same time.

Also AI companies will presumably store generated material long term so that any supposed evidence can be checked against AI company records.

I suspect law enforcement will ultimately have to use AI to scan if something has been made by AI and I assume that on a pixel level there would still be some inconsistencies that would occur from it having been generated and not recorded.

If it's impossible to determine if image/video evidence is real or not it'll simply be like how a large amount of evidence is already treated - as possibly ambiguous and suffering from reasonable doubt.

For example - a blood spot on someone's carpet from a murder victim.
Is it because the accused murdered them and failed to clean up 100% of the evidence?
Or is it because the murder victim was their friend and suffered a nosebleed a year earlier and it wasn't bleached out?

On a jury you have to weigh the balance of probability with all evidence today, the future will be pretty much the same.