Most AI systems aren't ready. Check yours in 15 min →
GO

Gemini Omni Video: How to Identify AI Generation Sources in Replies

AuthorAndrew
Published on:
Published in:AI

This is the kind of thing that looks like a harmless flex—until you realize what it trains everyone to accept.

A video is circulating on social media, and it’s been generated by something called “Gemini Omni.” The post isn’t subtle about it, either. People are pointing out that you can see the source of the generation in the first reply. So this isn’t a “gotcha” where someone tried to pass it off as real and got caught. It’s closer to: “Look what the machine can do now.”

And that’s exactly why it matters.

When synthetic video is clearly labeled, the instinct is to shrug. Cool demo. Funny clip. Scroll. But the real shift isn’t the video itself. The shift is how quickly our brains start to treat “a video” as just another format that can be produced on demand, like a meme template or a stock photo. The label becomes a footnote. The spectacle becomes the point. And then the next time the label isn’t there—or is cropped out, or buried—your mind is already warmed up to accept the reality-feel of it.

That’s the uncomfortable part: the best “ethical” demos are also practice rounds for the unethical ones.

From what’s been shared publicly, the clip is being used as evidence of capability. Not just “AI can make images,” but “AI can make scenes.” Motion, timing, tone. That’s a different animal. Images used to be the thing people argued about—Photoshop, filters, editing. Video used to feel like the courtroom evidence of the internet. “Show me the clip.” Now video is drifting into the same category as everything else: content that can be fabricated cleanly, quickly, and convincingly.

People will say: relax, you can tell it’s AI. Sometimes, sure. But “sometimes” is doing a lot of work there. Also, being able to tell doesn’t scale. You might have the media literacy to spot weird details. Your parents might not. Your group chat definitely won’t. And the whole point of these tools is they get better fast. Today’s “obvious” becomes tomorrow’s “pretty good,” and then it becomes “good enough that nobody checks.”

The consequences aren’t abstract. Imagine a local election where a clip drops the night before voting: a candidate “caught” saying something awful. Even if it’s debunked the next day, the damage is done. People don’t update their emotions as quickly as they update their facts. Or imagine a workplace situation: a manager gets sent a video of an employee “admitting” to leaking something. HR doesn’t have time to play detective. They just want the problem gone. Who pays the price for that speed? Usually the person with less power.

There’s also a quieter consequence that I think is worse: we start treating real video as suspect by default. That sounds like a defense—“don’t believe everything you see.” But it cuts both ways. If everything can be fake, then anything can be dismissed. The guilty get a new line: “That’s AI.” The powerful get plausible deniability on tap. And regular people lose one of the few tools they had to prove what happened to them.

This is where I’m not neutral: I don’t think “label it” solves it. Labels help a little, but labels are social norms, not physical laws. They can be removed. They can be ignored. They can be faked too. And the incentives are obvious. If a synthetic clip helps your side win an argument, get views, or make money, the temptation to “forget” the label will be constant.

Now, I can hear the pushback: synthetic video also has real upside. Artists can make scenes they could never afford to shoot. Small businesses can create ads without hiring a full team. A teacher can generate a quick visual explanation instead of hunting for the perfect clip. Even normal people can tell stories, make jokes, or bring ideas to life. That’s all true. And I don’t want a world where only big studios and big budgets get to make compelling media.

But the uncomfortable truth is that the same feature that makes this empowering—the ability to create convincing video without friction—is what makes it dangerous. The cost of producing “evidence” is dropping. The cost of causing confusion is dropping. The cost of harassment is dropping. If you can generate a video of someone doing something humiliating, the harm doesn’t require anyone to believe it fully. It just needs to spread enough to stick.

What I don’t know is where the social breaking point is. Maybe people adapt faster than I expect. Maybe we build a new set of instincts: we demand context, we ask for originals, we slow down. Or maybe we do what we usually do online: we pick the version that fits our mood and call it truth.

Right now, a labeled “Gemini Omni” video is being shared as a demo, and that’s fine on the surface. But it’s also the normalization step. It’s the moment where synthetic video stops being a weird edge case and becomes “content.”

If we’re walking into a world where any clip could be generated, what do you think should count as proof anymore?

Frequently asked questions

What is AI agent governance?

AI agent governance is the set of policies, controls, and monitoring systems that ensure autonomous AI agents behave safely, comply with regulations, and remain auditable. It covers decision logging, policy enforcement, access controls, and incident response for AI systems that act on behalf of a business.

Does the EU AI Act apply to my company?

The EU AI Act applies to any organisation that develops, deploys, or uses AI systems in the EU, regardless of where the company is headquartered. High-risk AI systems face strict obligations starting 2 August 2026, including risk management, data governance, transparency, human oversight, and conformity assessments.

How do I test an AI agent for security vulnerabilities?

AI agent security testing evaluates agents for prompt injection, data exfiltration, policy bypass, jailbreaks, and compliance violations. Talan.tech's Talantir platform runs 500+ automated test scenarios across 11 categories and produces a certified security score with remediation guidance.

Where should I start with AI governance?

Start with a free AI Readiness Assessment to benchmark your current maturity across 10 dimensions (strategy, data, security, compliance, operations, and more). The assessment takes about 15 minutes and produces a prioritised roadmap you can act on immediately.

Ready to secure and govern your AI agents?

Start with a free AI Readiness Assessment to benchmark your maturity across 10 dimensions, or dive into the product that solves your specific problem.