A first look at Sora 2, OpenAI’s wild new video platform

When GPT5 was released a couple of months ago, it felt like a letdown. Where was the staggering new capability, the “wow” moment that knocked everyone for six? Perhaps, many wondered, AI was finally starting to slow down? 

Fear not (or stay afraid) — the pace of change is still rapid and the jaw-dropping moment has arrived.

OpenAI’s Sora 2 is a reality-warping tool with vast potential for creativity and chaos. Both a groundbreaking new video model that surpasses anything seen before, and a TikTok-style social media platform that allows you to immediately share your creations, Sora 2 is a classic, horizon-expanding, fear-inducing AI moment. 

Don’t believe me? Here’s a viral video of an American influencer getting his comeuppance with a cranky cassowary. This took about 10 minutes to generate – I needed to iterate the prompt several times to get it right, but this final version is impressive. I’d like to think that I’d spot it as a fake … but maybe I wouldn’t ? 

Initial opinions about the impact of Sora 2 are split, with some hailing it as “revolutionary” and others dismissing it as “AI slop Tiktok.” Access is currently by invitation only, and these invites are in hot demand. But after an hour of frantic work, I managed to snag one by lurking on OpenAI’s Discord channel. I use Android, so there’s no app for me, but I was able to access it through the sora.chatgpt.com website via a VPN (it’s only available in the US and Canada right now). 

My take? Marketers and publishers who dismiss Sora 2 are in danger of repeating the mistakes of the past. But the insane capabilities of this tool, along with OpenAI’s norm-shattering rollout, also foreshadow a whole lot of problems. 

The company knows it’s taking a risk — when you first login to the tool you receive this message: “You’re entering an AI-generated world. Faces may be familiar, but nothing is real.”

Nice warning, but downloading the videos out of this AI generated world is core functionality, and it won’t take much work to get rid of the Sora watermarks. If the age of the deep fake wasn’t on us before, it definitely is now.

A classic case of disruption

The “AI slop TikTok” label reminds me a lot of past dismissals: “YouTube is full of cat videos,” “no one will want to watch movies on a phone,” and “vertical video is a terrible aspect ratio”. All of these turned out to be classic disruption in action. A low-cost, democratising technology emerges that is in some ways inferior to existing options, and incumbents dismiss it because they can do things better.

But then this cheap new technology begins to chip away at their market share as young, eager creators dive in, discovering novel approaches and formats native to the new medium. 

Examples of this abound, but we saw this exact thing 15 years ago in the news industry, when we noticed that many of our users preferred raw video on phone cameras to fully produced packages from the 6pm news. Senior television people scoffed at this — why wouldn’t we use the fully produced package? People couldn’t possibly prefer this shaky, hand-held footage to the artfully produced work put out by their skilled camera operators, journalists, editors and producers. 

Look where we are now — many people regard carefully edited videos presented by well-groomed reporters as stale, lacking authenticity and even as untrustworthy. 

With Sora 2, this transformation feels like it could happen faster than ever. Social feeds are already flooded with its creations. The video quality represents a generational leap, surpassing Google’s Veo 3, which felt like a breakthrough just months ago. Physics are more accurate, the model can perform edits, and character consistency is impressive. Perhaps most remarkable is the sound—it’s rich, synchronized, and natural.

Just listen to this ASMR video I made featuring a creepy Winston Churchill. 

Packaging this technology into a TikTok-style app is a clever move. It really leans into the AI-ness of the video, rather than trying to exactly replicate the types of outputs that come from a camera. I think that’s an essential point — the uniqueness and weirdness is a feature, not a bug, and the best creators will learn to work with that. 

Personally, I haven’t enjoyed a new product this much since, well, I can’t really remember. Maybe when I got Thunderpunch He-Man for my 10th birthday? I think Sora 2 is likely to capture a slice of the audience from TikTok and Instagram in the vertical video space.

Mind-bending the guardrails

But the launch raises so many questions. Will the rampant copyright theft get the app shut down within a week? How long until a brand or public figure gets pulled into some sort of public issue from a Sora-generated video? How long will OpenAI keep providing very generous free access to this undoubtedly expensive model (I haven’t run out after using it all day)? And is another “clickhole” really something the world needs? 

OpenAI once declared its goal was to “advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return”. Sora 2 suggests that the need to generate a financial return is now top-of-mind, and eating Tiktok’s lunch is the best way to get it.

The potential disruption from a video model this good is mind-bending. So much of our reality is mediated by video. We’ve been concerned about deepfakes for a while, but this may be the technology that pushes it into overdrive, to the point where we just stop trusting every video we see online. Even if OpenAI implements reasonably effective guardrails (and they do have some), other models will inevitably catch up. Some will explicitly market unmoderated capabilities if there’s no regulatory pressure.

One of Sora 2’s key features (which I wasn’t able to try out because I’m not using the iOS app) is Cameo, which allows users to insert their own likeness into videos. OpenAI chief Sam Altman has made his Cameo generally available and the largest number of deepfake videos on Sora involve him (including one of him shoplifting that’s gone viral).

But it’s not hard to get around the guardrails. Here’s a video I made of a reporter saying that Nvidia’s share price had plunged precipitously due to a shortage in rare earth minerals. It wouldn’t let me use CNN’s logo, but there were no issues when I asked it to use Marketwatch’s branding instead. Note that this video is not available for viewing unless you have this link and is very clearly marked as fake to ensure that this is not taken seriously by anyone:

Copyright and brand safety: A Wild West

It’s also clear that OpenAI is doubling down on its cavalier approach to copyright. In fact, the company is forcing copyright holders to actively opt-out of the model if they don’t want their IP being used for crazy Sora videos. 

Sora 2’s biggest stars are, for now, Pikachu, Spongebob, Mario, Princess Peach, and Peter Griffin. I made Astro Boy punch Mecha Spongebob in the face without any complaint. 

How long until this party gets shut down? This novel approach will surely be tested in court (although Disney, which is currently suing image AI company Midjourney, has already opted out). 

Brand safety is also non-existent. I prompted the Michelin Man to twerk suggestively in a rubber cell while making ASMR sounds. It’s pretty icky. I’m sure the folks at Michelin will be thrilled.  

I also made this video of a boy finding a scab in his corn flakes. It’s not incredible, but that Corn Flakes box sure is prominent.

Not perfect, but astoundingly fast progress

Sora 2 has flaws. There’s often something slightly glitchy or “off” in these videos. But considering that at the beginning of this year we were all laughing at weirdly morphing gymnasts and hideous AI mutations, the pace of change is astoundingly fast. There’s a solid chance that Google or another challenger will respond with an even more advanced model soon. Those glitches will start to disappear.

To sum up: I believe Sora 2 is a viable social app with the potential to produce countless viral videos, and creators will find innovative ways to use the technology. I think ignoring it or pretending that it’s “just slop” is a stupid idea. And I think OpenAI will make it better with new features (like image-to-video, more granular control over outputs and studio-like editing features). In the short term, I’m most interested in whether the rampant copyright theft will lead to it being temporarily shut down by a court, and whether its current trajectory of massive growth will sustain beyond the initial novelty stage.

Personally, I think this monster is here to stay. 

Related Content

Features Nano Banana: The AI that’s ripe for image editing

F.Y.I. Shaun Davies announces 'The AI Training Company' venture