I recently tested Google Gemini’s newest, much-hyped video generation model, Veo 3. Part of Gemini’s extremely expensive $250-per-month AI Ultra plan, Veo 3 can render small, finely detailed objects, like chopped onions, in motion and create accompanying, realistic audio. It isn’t perfect, but with some careful prompt calibration and enough generations, you can create something indistinguishable, at a glance, from reality.
Yes, this is cool, deeply impressive new technology. But it’s also a lot more than that. It might mean the final death knell for truth on the internet. Veo 3 already poses a major threat as is, but just one minor update will revolutionize deepfake creation, online harassment, and the spread of misinformation.
Once Veo 3 Gets the Image Upload Feature, It’s All Over
For all the upgrades the Veo 3 model has over its predecessor, Veo 2, it’s currently missing a key feature: the ability to generate videos based on pictures you upload.
With Veo 2, I can upload a picture of myself, for example, and have it generate a video of me working on my computer. Considering that Veo 2 and Google’s AI animation tool, Whisk, both support this functionality, it seems inevitable that Veo 3 will get it eventually. (We’ve asked Google if it plans to add this feature and will update this article with its response.) This would mean that anybody will be able to generate lifelike videos of people they know doing and saying things they never have and probably never would.
The implications are obvious in an era where clips of dubious authenticity spread like wildfire on social media every day. Don’t like your boss? Send a clip to HR of them doing something inappropriate. Want to spread fake news? Post a faux press conference on Facebook. Hate your ex? Generate them doing something unseemly and send it to their entire family. The only real limits are your imagination and your morality.
If generating a video with audio of a real person takes only a few clicks and doesn’t cost much (or anything), how many people will abuse that feature? Even if it’s just a tiny minority of users, that still adds up to lots of potential for chaos.
Google Isn’t Serious About Moderation
As you might expect, Google imposes some limitations on what you can and can’t do with Gemini. However, the company isn’t nearly strict enough to stop the worst from happening.
Of all the chatbots I’ve tested from major tech companies, Google’s offering, Gemini, has the weakest restrictions. Gemini isn’t supposed to engage in hate speech, but it will give you examples if you ask. It isn’t supposed to generate sexualized content, but it will provide an image of somebody in beach attire or lingerie if you prompt it. It isn’t supposed to enable illegal activity, but it will create a list of the top torrenting sites if you so inquire. Basic restrictions for Gemini that prevent it from generating a video of a popular political figure just aren’t enough when it’s so easy to get around Google’s policies.
Get Our Best Stories!
Your Daily Dose of Our Top Tech News
By clicking Sign Me Up, you confirm you are 16+ and agree to our Terms of Use and Privacy Policy.
Thanks for signing up!
Your subscription has been confirmed. Keep an eye on your inbox!
(Credit: Reddit/PCMag)
What happens when Google’s already lax restrictions meet an internet community intent on breaking them? Take ChatGPTJailbreak, for example, which is in the top 2% of subreddits by size. This community dedicates itself to “unlocking an AI in conversation to get it to behave in ways it normally wouldn’t due to its built-in guardrails.” What will like-minded people do with Veo 3?
Recommended by Our Editors
I don’t care if someone wants to amuse themselves by getting a chatbot to generate adult content or rely on one for finding torrenting sites. But I am worried about what easy-to-generate, photorealistic videos (complete with audio) mean for harassment, misinformation, and public discourse.
How to Deal With Veo 3’s New Normal
For every SynthID AI content watermark system Google introduces, third-party watermark removal sites and online removal guides appear. For every chatbot with restrictions and safeguards, there’s a FreedomGPT without them. Even if Google locks Gemini down with so many filters that you can’t even generate a cute cat video, there’s very little in place to stop jailbreakers and uncensored imitators once Veo 3-style video generation becomes mainstream.
For decades, sketchy Photoshopped images depicting real people doing things they never did have made the rounds on the internet—these are just part of life in the digital age. Accordingly, you must fact-check anything you see online that seems too awful or too good to be true. This is the new normal with Veo 3 video generation: You can’t treat any video clip you see as the real thing, unless it’s from a reputable news organization or another third party you know you can trust.
Gemini’s Veo 3 video generation is just the first skip of a stone across the pond of widely accessible, truly lifelike AI video generation, too. AI video generation models are only going to get more realistic, offer more features, and proliferate more, too. Gone are the days when video evidence of something is the smoking gun. If truth isn’t dead, it’s different now and requires careful verification.
About Ruben Circelli
Analyst, Software
Leave a Comment
Your email address will not be published. Required fields are marked *