Elon Musk’s so-called “free speech AI” just got a new update, and it could quickly turn into a tool for celebrity deepfakes or political turmoil. The company just announced Grok Imagine v0.9, which brings several improvements, including support for sound, allowing users to add custom speech in videos.
Along with that, Musk has made the video generation tool free for everyone. But it’s the last two aspects that can land the company in serious legal trouble unless proper guardrails are put in place.
Grok Imagine Goes Fully Public to Take On Sora 2
The new Grok version 0.9 update is a significant one, especially for its Imagine video generation model. The older version was incapable of adding sound to videos, and it was limited to X Premium and SuperGrok users only.
But the latest update, these issues are being addressed, along with improvements to video quality, going from 16 FPS to a smoother 24 FPS. Elon Musk announced the release of this new update in his X post.
This seems like a direct response to OpenAI’s Sora 2, which launched a couple of days back, and has an invite-only system limiting users who can use the new video generation tool. Grok Imagine is also quicker when it comes to generating videos, taking mere seconds in comparison to Sora, which could need a minute or two.
I Used Grok Imagine to Make Trump Say Anything I Wanted
Given the hype on X, I wanted to try out what’s really new with Grok Imagine and what all it’s capable of making. To start with, I tested casual prompts like 2 animated animals talking to each other, a Tesla racing F1 cars, a jolly dog, and a car made out of bubble wrap. Everything looked pretty neat so far.
Later, I asked it to create an Ice Princess, and used the new Speech option from the drop-down menu to say a few corny things. It pulled it off quite well. Given the controversial nature of Grok Imagine, I was curious to test how strict its guardrails really are. So, I typed some curse words, and the princess delivered every line without once refusing to do so.
I wondered if it could do that, would it allow me to do the same with real-world public figures? Could I create a video of political leaders say things they never said? I did not have to wonder for long, as Grok fulfilled my request without a hitch. I was able to create videos of President Donald Trump and made him say random lines, since there is no restriction whatsoever.
The speech, just like the visuals, was almost convincing. It was realistic enough that a low-res video could easily pass as genuine on Facebook or X. I created a video of Musk in a far-right rally saying all the not-so-good things on stage, and not once did Grok refuse to take my requests, something ChatGPT or Gemini would have outrightly refused. Grok also didn’t refuse to use the likeness of public figures.
Free Grok Imagine is a Deepfake Machine with Zero Guardrails
If that were not enough, Grok’s controversial Spicy mode is here for all the “other” types of requests. The fact that you can generate images that share a close likeness to celebrities is already highly concerning, but the Spicy option goes a step further and those same figures can be rendered in explicit or suggestive poses within seconds.
It renders the person in the video almost completely undressed, with only their undergarments on, and exaggerated body features. It’s honestly downright absurd how easy Grok makes it to create deepfakes, characters willing to say whatever depraved things you want them to say.
I understand Musk’s desire to keep Grok as a platform with minimal restrictions compared to the likes of Gemini and ChatGPT, but some guardrails are still essential. Without safety checks, I can easily imagine my X feed flooding with deepfakes of popular celebrities and prominent figures saying radical things that they haven’t said, thanks to Musk’s Grok Imagine.
