Here's the good news: Omni, which was unveiled this week at Google I/O 2026, is not fast. Now the bad news: its output is insanely good. It'll be hard to tell what's hand-crafted and what's been generated by the multi-talented Gemini Omni.
I guess that's where SynthID comes in. It's now part of the Gemini app and can assist in image, audio, and video verification.
"You can learn about an image's origin by using Search features like Lens, AI Mode and Circle to Search, as well as Gemini in Chrome. Just ask, 'Is this made with AI?' or 'Is this AI generated?'" wrote Laurie Richardson, Google's VP of Trust and Safety, in a blog post detailing all the SythID updates.
But the irony here is as thick as the San Francisco fog rolling in behind the Golden Gate Bridge: lovely to look at, but also impenetrable.
The very same companies building ever-better models for image, video, and audio generation are also creating the tools that let us invisibly watermark and/or identify AI-generated content.
Imagine Peanut's Pig Pen character trying to use a vacuum on his dust cloud, and you get the idea. Pig Pen could take a bath, but then what would be the point? He has no other name and must remain dirty.
Google, OpenAI, Eleven Labs, and others partnering on SynthID can no more stop improving their models and output than Pig Pen can clean himself up. It's not in their DNA.
The Gemini is out of the bottle
It's not that I don't appreciate the effort. Watermarking AI content is about the only way for us to catch and remove fake images, audio, and video. We know we can no longer trust our eyes and ears, and most of us have been fooled at least once by really good AI.
With models like Omni, those instances are sure to rise, so what else can these companies do but find a way to help us spot generative content?
But doing that, and announcing those efforts in the same breath as they unveil the mind-bending Omni, which seems able to create almost anything from any input, is disingenuous, to put it lightly.
It's like they don't realize they're talking about themselves when they say, "As generative media becomes more advanced and accessible, it’s helpful to know where content comes from, and whether it’s been altered." It comes from you.
Of course, the language here matters, and, notably, they never use the term "fake content" when talking about AI.
Google certainly acknowledges that people are using their models to create generative content, touting its use of C2PA Content Credentials (useful for identifying the provenance of generative and non-generative content), but I don't think it does much to acknowledge that companies like it and OpenAI contribute to our free-floating anxiety when struggling to identify what is real.
The allure of the unreal
On the one hand, I love powerful models like Omni. I started playing with it today, asking it to create a pair of videos. For one, I fed it a 6-second video of myself and asked it to, among other things, melt me. That arrived in under a minute, though the melt was less than complete. The second is a claymation video on the creation of the combustion engines. It's taking the better part of an hour.
Still, it already feels like a limitless creative tool for breaking down the barrier (lack of artistic or coding skill) between your good idea and output. The power of AI is that you only need a really good prompt to create something capable of fooling the average human.
As I said, I'm glad SynthID exists, but I realize now it doesn't belong in one company's hands. Google may have created it, but a third party, preferably one outside the AI industry and connected to intellectual property, content verification, and maybe digital media content (possibly a news organization), should probably control its use and development. They might offer a view unavailable to Google and its partners, who are likely more focused on the next best AI model and not on trustworthy content.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.