In addition to the cryptography, digital trace, and secure storage/chain of custody answers already provided, also consider the possibility of statiscial analysis that can identify manipulation in subtle ways.
Here is a link to an article that describes some of these analyses:
https://articles.forensicfocus.com/2013/08/22/detecting-forged-altered-images/
This article discusses how simply opening and saving a file can change the structure of a picture without producing visual differences; how inserting a new image into another picture will leave mathematical artifacts that can be discovered with error analysis; and how subtle differences in quality in different portions of the same photo can be proof that the image was manipulated.
Now, there may be techniques that can be used that might mitigate some of this: one idea that comes to mind is to perfectly model a scene, and then create a perfect CGI with no artifacts. Even in a case like this, however, it may be possible to show statistically that the scene is fake: digital algorithms themselves often leave certain statistical "markings" that might be identifiable -- and, for that matter, going back to the quality issue, such a generated scene might have a quality that doesn't match the available video equipment that allegedly filmed the incident (ie, grainy when the camera would have had a high amount of detail, or highly-detailed when the camera would have been grainy).
Finally, it would be important to collaborate video evidence with details from the real world. A forger, for example, may have known that the individual in question wore a popular designer jacket, but would not have been aware of the tear in the jacket that happened at a party the night before, in a spectacularly embarrassing incident that everyone saw -- or, alternatively, that photo that the prosecution is claiming to be fake also happens to show a name-tag sewn on the jacket, that the person alleged to have faked the photo would have no reason to know actually existed...