OpenAI · Sora 2 · AI Video
Sora 2 was one of OpenAI’s most ambitious experiments in generative video. Introduced in September 2025 alongside a social app built around AI-generated content, it promised greater physical realism, synchronized audio, and new ways to place real people inside synthetic scenes. In April 2026, OpenAI discontinued the Sora experiences on the web and in the app. Its short lifespan, however, left several lessons about where AI video could evolve next.
When OpenAI introduced Sora 2 on September 30, 2025, the conversation around generative artificial intelligence was entering a new stage.
Generating images from text instructions had already become commonplace. The next challenge was to achieve higher levels of visual consistency, motion, control, and sound in video.
Sora 2 attempted to move precisely in that direction.
But OpenAI did something else: it did not simply introduce a model for generating videos. It also launched a social experience built around those synthetic videos.
The combination made Sora 2 a particularly interesting experiment.
Less than seven months later, OpenAI discontinued the Sora experiences on the web and in the app.
Knowing how the story ended allows us to analyze the project from a different perspective: not only for what it promised, but for what it actually represented in the evolution of AI-generated video.
What Was Sora 2?
Sora 2 was a video and audio generation model introduced by OpenAI as the successor to the first generation of Sora.
OpenAI highlighted improvements in areas that remain fundamental to generative video models:
- greater physical realism;
- improved instruction following;
- greater control over scenes;
- synchronized audio;
- dialogue and sound effects;
- greater consistency in certain types of motion.
One of the ideas behind this evolution was for models to stop trying to satisfy a visual instruction at any cost.
OpenAI used the example of a basketball that fails to go through the hoop. In earlier models, the system could distort reality to make the ball end up inside. Sora 2 aimed to represent the physical consequence of the miss more accurately.
The advancement was not simply about generating better-looking images, but about making what happened from one frame to the next more coherent.
Sora 2 Also Generated Audio
Another important difference compared with earlier stages of audiovisual generation was the addition of sound.
Sora 2 could generate video accompanied by elements such as dialogue, ambience, and sound effects synchronized with the scene.
This brought the model closer to a more integrated form of audiovisual generation.
Instead of producing the visuals first and then relying on other tools to fully build the audio layer, the system could interpret both components as part of the same creation.
The importance of this change went beyond convenience.
Sound is part of audiovisual storytelling. A scene can change completely depending on the voices, silences, effects, and ambience that accompany it.
What Was the Sora App?
The launch of Sora 2 was accompanied by an app that proposed something different from a conventional video generation tool.
Users could create videos and discover content generated by other people within a feed.
This introduced a social dimension to the product.
The idea was not simply to write a prompt, download the result, and leave the tool.
OpenAI was experimenting with a space in which creating, discovering, reusing, and sharing synthetic content could all be part of the same experience.
This decision led to comparisons with social video platforms, although there was one fundamental difference: Sora’s content was built around AI generation.
Cameos: Placing Real People Inside AI-Generated Videos
One of the app’s most striking features was Cameos.
The feature allowed a person to verify their identity through a video and audio recording and then use their likeness within generated scenes.
They could also authorize other people to use that representation.
This introduced a powerful idea: a person’s identity could become a reusable element within a synthetic production.
But it also raised important questions related to:
- consent;
- identity control;
- use of a person’s likeness;
- impersonation;
- revocation of permissions;
- content authenticity.
Sora 2 Anticipated an Increasingly Important Debate: Who Controls Our Digital Identity?
The ability to place a realistic representation of a person inside AI-generated videos changes the traditional relationship between identity and audiovisual production.
Previously, appearing convincingly in a scene required being physically in front of a camera or undergoing a complex visual-effects process afterward.
Generative models are progressively reducing that barrier.
This creates creative possibilities, but it also makes consent a fundamental part of the technology.
It is not enough to ask whether a model can reproduce a face or a voice.
We also need to ask:
who authorized that use, for how long, and who retains control over that representation?
Was Sora 2 a Social Network?
Describing Sora simply as an “OpenAI social network” may be too broad.
The app did incorporate features associated with a social platform: a feed, discovery, creation, interaction, and content reuse.
But its primary purpose was closely tied to experimenting with audiovisual creation through Sora 2.
That is why it is more accurate to understand it as a social experience built around generative video.
The distinction matters because it avoids assuming that OpenAI was necessarily building a conventional competitor to Instagram or TikTok.
The Experiment of a Feed Made Up of Synthetic Content
One of the most interesting aspects of Sora was the question of what happens when a social platform uses artificial intelligence not only to recommend content, but also to produce it.
On traditional social networks, people create videos and algorithms decide which ones to show.
A generative platform can add another layer:
artificial intelligence participates in both creation and distribution.
This opens up creative possibilities, but it also raises questions about saturation, authenticity, and value.
If producing a video takes only a few minutes, the volume of available content can grow enormously.
The question then stops being simply who can create.
It becomes what content deserves our attention when creation itself is no longer scarce.
When Was Sora 2 Discontinued?
OpenAI discontinued the Sora experiences on the web and in the app on April 26, 2026.
This means that the Sora 2 experience introduced to the public in 2025 is no longer available today in the same form.
The Sora 2 API remained temporarily available to developers, with a shutdown date announced for September 24, 2026.
Therefore, this article should be understood as a retrospective analysis of a technology and product that belonged to a specific stage in the evolution of generative video.
Why Did OpenAI Discontinue Sora 2?
Here, it is important to separate facts from interpretation.
OpenAI confirmed the discontinuation and announced the corresponding dates, but that does not mean we should attribute the shutdown to a single cause without sufficient official evidence.
Artificial intelligence products evolve rapidly, and companies may retire models, interfaces, or entire experiences as their technology and product priorities change.
That is why it would be incorrect to claim that Sora 2 disappeared simply because it “failed,” because it had too few users, or for any other reason that OpenAI did not establish as the cause.
The fact that a product was discontinued tells us what happened. It does not necessarily explain, by itself, why it happened.
Sora 2’s Discontinuation Does Not Erase Its Importance
In technology, longevity and influence are not always the same thing.
A product can disappear and still have served as a testing ground for ideas that later emerge in other tools, models, or experiences.
Sora 2 brought together several trends that remain relevant:
- joint video and audio generation;
- greater physical consistency;
- reusable identity through AI;
- consent over digital representations;
- audiovisual creation accessible through natural language;
- social distribution of synthetic content;
- provenance and authenticity of generated media.
That combination of ideas makes Sora 2 an interesting case even after its discontinuation.
Sora 2 and the Problem of Authenticity
As AI-generated video becomes more realistic, it becomes increasingly important to distinguish between visual quality and authenticity.
A video can look completely convincing and still depict a synthetic scene.
This is not a problem when the nature of the content is clear.
The risk arises when a generated piece is deliberately presented as evidence of a real event.
That is why the evolution of generative video is closely connected to debates around provenance, Content Credentials, C2PA, and media literacy.
The ability to generate convincing images needs to evolve alongside mechanisms that help us understand where content comes from.
What Did Sora 2 Leave Behind for Audiovisual Professionals?
For those working in audiovisual production, Sora 2 was another sign that artificial intelligence was reducing some of the technical barriers to creation.
A person could describe a scene in natural language that traditionally might have required cameras, locations, actors, lighting, sound, visual effects, or animation.
But reducing technical difficulty does not eliminate the need for creative decisions.
A tool can generate images, movement, and sounds.
We still need to decide:
- what we want to communicate;
- what story we are telling;
- what shot we need;
- what the audience should feel;
- what pacing the piece requires;
- which elements need to remain consistent;
- when to use AI and when traditional production delivers better results.
When generating images becomes easier, the judgment required to decide which images we actually need becomes even more valuable.
Generating a Clip Is Not the Same as Directing a Production
This distinction is especially important when discussing artificial intelligence applied to video.
Getting a visually impressive clip from a prompt is not necessarily the same as building a coherent audiovisual piece.
A production requires continuity, narrative intention, consistency across characters and objects, editing, sound, and control over multiple decisions.
Generative models can participate in that process, but the final result still depends on how they are integrated into a workflow.
Professional value therefore shifts from simply “knowing how to operate a tool” toward knowing how to direct different tools to achieve a specific result.
What Did Sora’s Social Experiment Teach Us?
Sora also left behind a question that remains open:
do we want social networks where a significant share of the content we consume is artificially generated?
The answer will probably depend less on the technical ability to produce those videos and more on the value people find in them.
A platform can generate virtually unlimited amounts of content.
Human attention, however, remains limited.
This means that an abundance of production does not eliminate the need for relevance, creativity, intention, and connection.
From Sora 2 to the Next Stage of AI Video
The story of Sora 2 illustrates how quickly artificial intelligence is evolving today.
In September 2025, it represented one of OpenAI’s most ambitious bets on audiovisual generation.
By April 2026, the web experience and app had been discontinued.
That short cycle should also serve as a reminder for professionals and companies:
it is not wise to build a strategy around a specific tool as if it were permanent.
Models, platforms, and products can change, become integrated into other systems, or disappear.
The capabilities we develop to use them—storytelling, direction, communication, visual thinking, and judgment—are far more transferable.
Sora 2’s Legacy Does Not Depend on It Still Being Available
Sora 2 does not need to remain active to continue being relevant as a case study.
For a few months, it brought together several of the questions that still surround AI-generated video:
How far can realism go? How do we incorporate sound? How do we protect our identity? What happens when anyone can appear in a synthetic scene? How do we identify where content comes from? And what value will creation have when generating images is no longer difficult?
Those questions outlived the product.
Tools can disappear. The transformations they set in motion can continue long afterward.
That is why Sora 2 deserves a place in the recent history of AI video: not as a tool we can still use, but as an experiment that simultaneously revealed the potential, tensions, and pace of change in generative audiovisual production.