Sora 2 Can’t Do Everything, But Damn Can It Do A Lot
OpenAIs new text to video / audio tool signals both opportunity and obsolescence
If I were a teen right now dreaming about filmmaking, Sora 2 would be my obsession. I’d be desperate for one of those coveted invites that opens the app up, having already cut my teeth on Veo 3, Kling, and Sora 1. I’d be amazed at what this latest text-to-video platform does so damn well.
The video below showcases some of the exceptional work that users have produced. This is only pre-launch - OpenAI has made Sora 2 available to about 1000 people, stress-testing it before rolling it out to the masses. Checking out the stunning and often surreal videos these early adopters are creating is gobsmacking. The quality. The fidelity. The sheer realness of fantastical worlds.
One of AI’s chief problems is that it’s neither embodied nor world-wise. It doesn’t know what it’s like to move through the world, nor does it understand how the world actually works. Our bodies give us an intuitive understanding of what reaching, walking, and running look and feel like from the inside. Our world-sense is our intuitive grasp of how other bodies, animals, and objects behave on the outside. We get what sunlight dancing on waves or a log splintering under an axe looks like.
AI has to be taught all of this. It has to learn what we grasp instinctively, be it inside or outsdie.
What’s easy for us takes enormous work when a GPT transforms words into imagery. Get it wrong and it looks fake. Get it slightly wrong and something feels off. Get it right? You see a world that behaves the way you expect it to.
When I look at many of these Sora 2 clips, no matter how fantastic, that’s what I see. Sora 2 has gotten more of these things right than Sora 1 or, for that matter, many of its competitors. (At least until a rival firm releases their next model.)
Monsters, pop culture characters, even famous people can be inserted into Sora 2 and come out the other side looking real.... or just real-ish or real-er. It’s no wonder Sora 2 became one of the most popular apps in the App Store, with invites becoming highly sought-after commodities and sources of fraud. (They have added tools to prevent unauthorized deepfakes. Let’s see how that works.)
By OpenAI’s own admission (and third-party testing), Sora 2 still struggles with things like multiple characters speaking, character consistency, and audio across montages - things we take for granted. But in 20-second chunks (the max Sora 2 allows), most people won’t notice the subtle discrepancies. The clips are too short, the motion too fast. Big discrepancies are a different matter.
Adult me, watching this unfold, gets that Sora 2 isn’t going to replace Hollywood overnight. Twenty-second videos aren’t long-form content. You can’t type a prompt and have a fully rendered movie appear moments later. While that might be possible one day, that all-or-nothing frame misses the point.
Things only need to be good enough to change everything in significant, often unforeseen ways.
What few saw coming was OpenAI creating a showcase for Sora 2 work - essentially their own TikTok or Meta Reels. They’re building a social media platform around their content, a dedicated space for people (especially the economically precious young demographic) to display and influence what gets made.
As Sora 2 reaches more users, they’re growing a community around their product. Technically, this is brilliant - the platform enables OpenAI to gather data on every watch, developing algorithms to display work and improve the AI that creates it.
Oh, and it does something else: it creates a space to monetize.
Money, Money, Money
AI platforms are desperate for revenue streams beyond pro fees, developer subscriptions, and licensing agreements. By building out a social media space, the user base - creators and viewers - all become income sources. Lots and lots of income. TikTok generated over $23 billion in revenue in 2024. OpenAI, which hasn’t found a strong monetization strategy, might have just discovered one.
Sure, much of what Sora 2 produces will be AI slop. But there’ll be excellent work too, stuff that can easily go viral. (And don’t forget - slop only has to be “good enough” to have immediate value.)
For inexperienced creators, Sora 2 is a gateway into the media space. It lets you experiment at the intersection of story, visuals, and sound. It lets you learn the language of these elements or express what you’ve already learned. But there’s a larger language to storytelling. While AI incorporates some of that into its models, it doesn’t understand that language the way humans do. It takes human understanding to do interesting things with this technology.
If you’re experienced, if you’ve been working as a creative for a while, you can’t see these tools as the end. Yes, there’s plenty to worry about. But there’s also tremendous opportunity if you’re inclined. Artists already know how to use embodiment and world-sense in their art - how to move the camera, frame an image, alter color to evoke specific feelings. Sora can ape some of that, but it doesn’t know how to do it purposefully or powerfully.
Not the Tool but the Tool-User
The Adventures of Reemo Green shows what you can do when you understand how to use these tools to create, not just throw out prompts and see what sticks. The sound, effects, story, and jokes in the short all work. That’s not something the tool can do on its own.
Look deeper at Reemo Green and you’ll see multiple tools involved in its creation. It wasn’t just typing in a prompt - it required using prompts and ideas across several applications, plus editing skills to make footage work together as a story should. That’s the product of experience. Sora 1 was used in creation, along with Runway Act Two, Udio, Nano Banana, Kling, Veo 3, ElevenLabs, Suno, Photoshop, and Premiere to edit all the pieces together. One person with talent and insight pulled it all together - but it required mastery across multiple platforms.
A less involved but still valuable application is for experienced creators to make trailer-style spec demos of larger work -the kind of thing that helps sell an idea. Experienced actors can insert themselves into digital calling cards. What would have taken thousands of dollars and days to film can now be iterated on in an afternoon.
The downside? Less money spent means less work for others and more technical skill required by everyone. On a larger scale, this tech will continue making inroads into the wider industry, reducing crew contributions - especially in VFX - significantly. No, you can’t type a prompt and get a full movie. Sora doesn’t easily understand shot differences. But the technology is evolving so those who do understand can create those kinds of things within this and other apps.
All of this has happened before….
OpenAI intially announced that rights holders will need to inform them if they don’t want their IP used in the product. Sound familiar? If you look at early Sora 2 content, there’s copyrighted work everywhere. YouTube famously built its early business on uploaded copyrighted content, only implementing blocking systems after the business matured. Sam Altman has since walked that statement back — and the more cyncial have suggested all of this might have been reconsidered or used for publicity.
This does nothing about rights issues for everyday people whose work trained these models. There’s still source material - performances and scenes forming the raw material the AI is built on. That’s still an unresolved issue.
Along those lines, Anthropic recently agreed to pay $3,000 per work to authors whose writing trained their models - if they can demonstrate their material was used. What about video creators whose work trained Sora? Given how Sora’s taken off (it’s one of the hottest App Store downloads right now), there will be money to go around. Just like Anthropic’s settlement, with the amount of investment capital and potential advertiser revenue coming in, there will be big PR plays for whatever modest amounts OpenAI pays for “confirmed” violations. Those payouts aren’t financial drains. They’re known costs of doing business.
Looking wider, concerns about job erosion in the professional community remain valid. This technology - remember, AI is a suite of technologies, not one integrated monster coming for us all - allows companies to do more with less. Elements of jobs will be chipped away. You’ll still need people in the loop, just fewer of them doing fewer things.
Will digital actors replace real ones? Sora 2 can’t create full films from scratch. (And you can’t encode whatever allows inspiration and experience to create astonishing performances.) But like digital video before it, this technology givestalented filmmakers access to tools previously exclusive to high-end productions. Think Windhorse (1998) and The Last Broadcast (1998) - the first indie films produced on prosumer equipment. The pattern repeats.
Kid, adult, or old-timer - there are opportunities and obsolescence with this text-to-video tech. It’s going to allow new creators to gain purchase, existing professionals to adapt to survive (if not thrive), and the usual corporate suspects to do what they do best: make gobs and gobs of money.