Can You Still Believe the Video You're Watching?

A fifteen-second video clip now costs about seven cents to generate with Google's latest video model, and the results are good enough to pass a quick glance. With that, a familiar worry comes back around: if anyone can fake convincing footage for pennies, can you believe any video at all?

Stated that strongly, the fear overshoots. But the version underneath it is worth taking seriously, and it is not quite the one people reach for first.

The case for alarm

The raw capability is real. Generation quality has climbed fast while the price has fallen, so short, realistic clips are now trivial to produce at scale. This is not hypothetical. AI-made political ads already ran during the 2026 midterms, and most of them did not disclose that they were synthetic. The tools are here and they are cheap.

Why "trust nothing" is the wrong lesson

The leap from "fakes are easy" to "video is worthless as evidence" skips over how people actually decide what to believe. A few things hold the line.

Provenance is being built into the pipeline. Google's generators stamp every clip with a SynthID watermark, and the industry-wide C2PA standard, often branded as Content Credentials, attaches tamper-evident information about where a file came from and how it was edited. Camera makers and large platforms have started adopting it. None of this is bulletproof, and watermarks can be stripped or degraded, so detection remains an arms race rather than a solved problem. But the idea that synthetic media travels with no fingerprints is already out of date.

Context still carries most of the weight. Consequential footage rarely arrives naked. It comes with a source, a publisher, a timestamp and other accounts that either corroborate it or do not. The clips that are genuinely hard to judge are the low-stakes-looking ones with no chain of custody, and those were often unreliable long before AI could make them.

The real cost

The sharper danger is not that you will be fooled by a fake. It is what researchers call the liar's dividend: once everyone knows realistic fakes exist, anyone caught on real video can wave it away as a fabrication. The threat to trust runs in both directions, and the second direction, genuine footage being dismissed, may do more damage than any single forgery.

So the honest answer sits between the two extremes. "Seeing is believing" was never literally true, and it is weaker now than it was a year ago. But video has not become worthless. Provenance, corroboration and source still decide what holds up, and the burden is less about spotting the fake than about refusing to be talked out of the real.