In a moment that blurs the boundary between self and spectacle, OpenAI has released Sora 2 — a tool that places ordinary people inside AI-generated worlds of their own imagining. Launching first in the United States and Canada, the application invites users to film themselves and then describe, in words, whatever reality they wish to inhabit. The release arrives not in a vacuum but amid deepening questions about who owns the images, stories, and likenesses that train and populate these synthetic worlds — questions that no demonstration video of a man backflipping on a paddleboard can fully ans
OpenAI's Sora 2 app lets users insert their faces into AI-generated videos
You film yourself, and the AI puts you anywhere you describe
Mark
So the basic idea is you film yourself, and then the AI puts you into any scenario you describe?
Mimi
Exactly. You record a short video of yourself with audio, and then you write a prompt—"me doing a backflip on a paddleboard," for instance—and the software generates that scene with your face in it.
Luke
How good is it at actually matching your face and movements? The source says some demo videos look realistic but others are bizarre. What's the actual quality bar here?
Mimi
The demos they showed are mixed. Some look convincingly real, some look surreal. The company is emphasizing that Sora 2 is better at complex movements like gymnastics, which suggests the previous version struggled with that.
Mark
And it's also a social network, right? You can see what other people made?
Mimi
Yes. You can view and edit other users' videos, and you can include friends in your own creations. It's designed to be collaborative.
Luke
The copyright thing is interesting. They're saying rights holders have to opt out rather than opt in. That's a pretty aggressive stance given how much pushback AI companies have gotten on this.
Mimi
Right. OpenAI is saying they'll include copyrighted material unless the studio or creator explicitly objects. But they're drawing a line at public figures—those require consent upfront.
Mark
Why the difference? Why is it okay to use copyrighted movies but not celebrity faces?
Luke
That's the question, isn't it. The source doesn't explain the reasoning. It could be legal strategy, or it could be that they think public figures have more standing to sue. But we don't actually know what OpenAI's thinking is here.
Mimi
What we do know is that entertainment companies have been complaining for years that AI was trained on their work without permission. This opt-out approach is essentially saying, "We're going to keep doing it unless you stop us."
Mark
And the app is only in the US and Canada right now?
Mimi
For now, yes. That's typical for a launch—they'll likely expand once they've worked out the kinks and legal questions.
O Pulso
- OpenAI has crossed a threshold: users can now film themselves and be transported by AI into any scenario they can describe, from moonlit landscapes to impossible athletic feats.
- The tool doubles as a social network, letting users watch, remix, and collaborate on each other's AI-generated videos — turning individual creation into a shared, evolving creative commons.
- Sora 2 shows marked improvement in rendering complex physical movement, with gymnastics sequences now generated with a convincingness that earlier models could not achieve.
- OpenAI's opt-out policy on copyrighted material has alarmed entertainment companies and creators, who argue their work is being used without consent unless they actively object.
- A single carve-out exists: generating videos that replicate the likenesses of public figures requires their explicit permission — though how OpenAI will enforce this at scale remains unanswered.
In a moment that blurs the boundary between self and spectacle, OpenAI has released Sora 2 — a tool that places ordinary people inside AI-generated worlds of their own imagining. Launching first in the United States and Canada, the application invites users to film themselves and then describe, in words, whatever reality they wish to inhabit. The release arrives not in a vacuum but amid deepening questions about who owns the images, stories, and likenesses that train and populate these synthetic worlds — questions that no demonstration video of a man backflipping on a paddleboard can fully answer.
OpenAI has launched Sora 2, an application that lets users film a short clip of themselves and then, through a text prompt, place their likeness into any AI-generated scenario they can imagine. The company demonstrated the tool with examples ranging from a man performing a backflip on a paddleboard to a figure skater balancing a cat on her head mid-triple axle — and even produced a clip of CEO Sam Altman walking through a virtual landscape and standing on the Moon.
Where earlier versions of Sora generated video from text alone, this iteration makes the user the subject of their own AI-produced scenes. Some results appear surreal and cinematic; others are convincingly lifelike. The app launches first in the United States and Canada.
Sora 2 also functions as a social platform. Users can watch videos made by others, edit them, and pull in friends who are registered on the service — transforming a standalone generator into something closer to a collaborative creative network. The update also brings notable improvements in rendering complex physical movement, particularly gymnastics sequences that require rapid, precise transitions between body positions.
The release lands amid significant scrutiny over intellectual property. Rather than seeking permission from rights holders before allowing their content to appear in generated videos, OpenAI is adopting an opt-out model — meaning studios and creators must actively object if they do not want their work used. This reversal of the traditional permission model has drawn criticism from entertainment companies who argue AI firms have long trained on copyrighted material without consent.
One boundary has been drawn: Sora will not replicate the likenesses of public figures without their explicit permission, creating a distinction between how the tool treats ordinary users and recognizable personalities. How OpenAI will verify or enforce that restriction at scale remains an open question — one that sits at the heart of a broader tension between enabling powerful creative tools and protecting the rights of those whose images and work make them possible.
OpenAI has released Sora 2, a new application that lets people insert themselves into artificial intelligence-generated videos. The tool works by having users film a short clip of themselves with audio, then using that footage to place their likeness into any scenario they describe through a text prompt. The company demonstrated the capability with examples like a man performing a backflip on a paddleboard and a figure skater executing a triple axle while balancing a cat on her head. The app is launching initially in the United States and Canada.
The software represents a significant step toward making AI video creation more personal and accessible. Where earlier versions of Sora could generate videos from text alone, this iteration allows users to become the subjects of their own AI-generated scenes. OpenAI's demo videos often show surreal or cinematic moments—movements and scenarios that don't quite match reality—but some clips appear convincingly lifelike. The company even created a video showing chief executive Sam Altman as a digitally rendered figure walking through a virtual landscape and standing on the Moon, illustrating the range of what the tool can produce.
Beyond individual creation, Sora 2 functions as a social platform for AI-generated video content. Users can watch clips made by others, edit them, and incorporate friends who are also registered on the platform into their own videos. This social layer transforms the tool from a standalone generator into something closer to a collaborative creative network, where people can build on each other's work and share results.
The second version of Sora shows particular improvement in handling complex physical movements. Gymnastics sequences, which require precise rendering of multiple body positions in rapid succession, are now generated more convincingly than before. The first Sora model launched last year; this update reflects OpenAI's continued refinement of how the software handles motion and spatial relationships.
The release comes as OpenAI faces ongoing scrutiny over how it handles intellectual property rights. According to reporting from The Wall Street Journal, the company is adopting an opt-out approach to copyrighted material and characters. Rather than seeking permission from rights holders like film studios before including their content in generated videos, OpenAI will allow such use unless the rights holder explicitly objects. This reverses the traditional permission model and has already drawn criticism from entertainment companies and creators who argue that AI firms have trained their systems on copyrighted work without consent.
There is one notable boundary: OpenAI has stated that Sora will not generate videos replicating the likenesses of public figures without their explicit permission. This consent requirement applies specifically to recognizable celebrities and public personalities, creating a distinction between how the software treats ordinary people and those with public profiles. The company has not detailed how it will verify consent or enforce this restriction at scale.
The Sora 2 launch reflects the broader tension in generative AI development between enabling powerful creative tools and protecting existing intellectual property and personal rights. OpenAI's approach—permissive by default for copyrighted material, restrictive for public figures—suggests the company believes the former concern is less urgent than the latter, or that rights holders have sufficient legal recourse to object. Whether that calculation holds as the tool spreads remains to be seen.
Citações Notáveis
OpenAI demonstrated the app's capabilities with a video featuring a digitally generated version of chief executive Sam Altman walking through a virtual world and standing on the Moon.— OpenAI
Quer a matéria completa?
Leia o original em The Star ↗
Technology