Key Takeaways:
- Generative video platforms remain an intellectual property no-man’s land, and the seemingly-successful launch of Sora, the new social AI video app from OpenAI, expands the creative toolkit without measurably advancing brand or creator guardrails.
- That app, along with this week’s go-live of the “instant checkout” feature in ChatGPT, is counter-positioned against TikTok, which has become a transaction-centric platform.
- The ability for Sora users to create “cameos” (the elements needed to consistently replicate people in video generations) of themselves, their friends, and celebrities, combined with the relatively open season on brand elements, could be about to usher in a very different model of social media trend generation.
At the same time TikTok (or at least its American operations) is potentially being hollowed out and separated from the algorithm that’s powered the most popular social media app of the last few years, OpenAI has chosen, probably not coincidentally, to launch its own confusingly-named Sora social AI video app.
As well as that being the designation for the video and audio generation model underlying the app, which is now in its second generation, Sora is also the name for the new consumer-facing product, which resembles nothing more closely than a version of TikTok where everything is produced by AI.
To hear OpenAI lead Sam Altman tell it, this new app is a way to help fund OpenAI’s longer-term ambition to build and fuel an unprecedented amount of data centre capacity to achieve its original not-for-profit ambition to create safe artificial general intelligence, but also a way for people to have some “fun” along the way.
Whether or not the product, the training data behind it, or even the top-level concept feel like “fun” to you (or whether that word trivialises a host of gigantic concerns) is going to be very dependent on where you stand on the inputs, outputs, and philosophical frameworks around AI. But what you can’t argue with is the surprising success: Sora is now the top “photo and video” app in the US App Store (it’s currently only available in America, and is also iOS-only) and the third most popular app across categories, eclipsing some previous big AI launches, but falling short of some others.
This success was by no means locked in. General-purpose AI competitor Meta launched its own AI video feed app, Vibes, at the end of September, to near-unanimous derision. It’s safe to say that Meta, in general, has not had a particularly good finger on the pulse when it comes to where the winds of culture are blowing over the last five years or so, but this felt like a particularly egregious case of what Ian Malcolm would have summarised as the “you were so busy thinking about whether you could, that you never stopped to ask whether you should” problem.
Or, to put it another way, it took an extraordinary amount of verve to ignore that backlash and put out a product that combines two of the most contentious things in consumer technology: generative AI, and algorithmic, engagement-based social feeds. But nevertheless, here we are: OpenAI now has two widely-downloaded apps, in Sora and ChatGPT, that seem to have captured the zeitgeist and steamrolled through a lot of the objections. And according to Sam Altman, Sora – the app – has been designed to, amongst other things, “optimise for long-term user satisfaction” and “prioritise creation”. So… TikTok with AI, as a concept, seems like it’ll be sticking around.
This is also, again not coincidentally, the week that OpenAI’s partnerships with eCommerce companies bore fruit, and the company added the “instant checkout” feature to ChatGPT. Because as much as TikTok built its name on passive media consumption, its latest turn has been very much driven by transactions – not least as a way for TikTok to turn its unequalled cultural relevance into per-user monetisation.
There is not, to be clear, any way to shop from within Sora right now. Nor is there any indication that there will be. But there are certainly a lot of signs pointing to the idea that there could be. And while some of them are direct, others are more roundabout indicators that what’s being positioned as harmless fun today could actually be a Trojan horse for a very different model of shopping.
Cameo is the feature that moves Sora from “AI video generator” into something resembling a different kind of social platform. Unlike other video generation platforms, Sora is built around the idea of users uploading their own likeness, or those of their friends (with differing levels of permission), and then using these in generations. Sam Altman has, himself, made his own cameo available, and the results are predictably cutting and compelling, from reasonably-believable-looking videos of the CEO stealing NVIDIA GPUs from an electronics store, to videos of Altman stealing drawings from Studio Ghibli’s Miyazaki while screaming “free art baby”.
Away from the comedic potential of a public AI figure opening themselves up to seemingly unfettered ridicule, this is also a reminder of just how much of an AI free-for-all the big generative models remain. In Sora 2’s first 24 hours, users generated clips of Pikachu storming the beaches of Normandy, Spongebob Squarepants being pulled over by police wearing a bodycam, and a cavalcade of other IP infringement.
Disney has already filed legal threats against Character.ai for the same issue. And generating brand-adjacent content, or even direct uplifts of brand materials, is as possible in OpenAI’s native image model as it was on day one, when users similarly waded straight into creating images crammed with celebrity likenesses and intellectual property.
But while the intellectual property side of things isn’t news, the cameo part of OpenAI’s vision could be, because it reframes what’s possible in the way people interact with brand IP in video generation models.
It may feel like a small change to go from a prompt asking for a model in a Balenciaga coat to a prompt asking for “me in a Balenciaga coat,” but the space between those things has the potential to actually be pretty vast.
Fashion already runs on compressed cycles and devolved creative power, where a single influencer look can spark global demand, and the addition of the personal, playful layer of cameo into a social app that could be posed for takeoff has the potential to speed that up further. And to ignore something like that would be to ignore potentially vital consumer intel, and with it the possibility of turning attention into revenue.
On the other side, once they engage with it, brands might find themselves stuck fielding requests for designs that were never meant to exist and that are impossible to actually realise. This is not new to anyone working in bespoke or made-to-measure fashion, where customers are already routinely coming through the door armed with generated images of styles that defy physics or the rules of patternmaking.
A lot of press this week has framed the launch of Sora (again, the app, not the model underneath it) as a litmus test for whether the user community will accept an AI social feed. And that framing ignores the fact that a lot of us already have: generated content is rife on other platforms, either through tacit permission, limited guardrails, or, in the case of Meta, explicit invitation. And those platforms are, by any reasonable definition, already pulling double-duty as fashion and beauty’s key storefronts and those industries’ inspiration feeds.
And that’s where fashion and beauty will need to pay attention, because the real question isn’t how convincing AI clips are becoming, or whether the world does decide that a feed full of them is “fun”. It’s a matter of whether users take up the mantle of inserting themselves into them and beginning to create in a playground where everyone’s IP and identity is up for grabs, and inspiration becomes a kaleidoscope of weirdness that will place a new burden on brands to separate what’s achievable and desirable from what’s not.