Sora

Sora was OpenAI’s attempt to push AI video generation beyond short visual experiments and toward a system capable of creating scenes that looked, moved, and behaved more like the real world. The original Sora model introduced text-to-video generation to a much wider audience, while Sora 2 later added stronger controllability, more accurate physical behavior, synchronized dialogue and sound effects, and a social creation experience built around remixing and shared characters.

There is one important detail anyone reading about Sora today needs to know: the product is no longer available. OpenAI discontinued the Sora web and app experiences on April 26, 2026, and the Sora API was discontinued on September 24, 2026. That means a modern Sora review is partly a review of what the product offered and partly an explanation of its current status.

That change does not make the technology irrelevant. Sora and Sora 2 remain important reference points in the development of generative video, particularly because they demonstrated how text, images, motion, sound, and scene consistency could be combined in one generation system.

Sora Started With a Bigger Idea Than Simple Text-to-Video

The original Sora was introduced as a text-to-video model designed to understand and simulate the physical world in motion. OpenAI’s early research described a system capable of generating videos from text while maintaining visual quality and following prompts. The research version was designed to produce videos up to a minute long, although the later public product had different practical duration and resolution limits.

The important concept was not merely turning words into moving pictures.

Sora attempted to represent what was happening across time.

If a prompt described a person walking through a scene, the model needed to maintain the person, surroundings, camera perspective, and movement from one moment to the next. When multiple objects appeared, the system had to keep their relationships reasonably consistent. These are difficult problems for generative video because each frame is part of a sequence rather than an isolated image.

That is why Sora was widely discussed as more than an ordinary animation tool.

Its research direction was connected to what OpenAI called world simulation: teaching models to represent aspects of physical environments and how those environments change over time.

The original product also gave users more direct creative controls than a single text box. When Sora became publicly available at sora.com in December 2024, OpenAI described capabilities including text generation, image and video inputs, Storyboard, Remix, Blend, and Loop. Users could also generate videos in widescreen, vertical, or square formats.

That combination made Sora interesting as a creative system rather than only a technical demonstration.

Sora 2 Took the Model in a More Realistic Direction

Sora 2 arrived in September 2025 as a new generation of OpenAI’s video and audio model.

OpenAI described it as more physically accurate, more realistic, and more controllable than earlier versions. One of the major additions was synchronized audio, including dialogue and sound effects.

That changed the nature of the output.

Earlier AI video workflows often required creators to generate silent visuals first and then add music, narration, or sound effects separately. Sora 2 aimed to generate the audio relationship alongside the video.

A person speaking in a scene could have dialogue synchronized with the action. Environmental sounds could form part of the generated sequence. The soundtrack was no longer necessarily a separate post-production step.

That does not mean every generated scene behaved like professionally produced film audio. Generative systems still had limitations, and complex interactions could fail. But synchronized audio made the generation process considerably more complete.

Sora 2 also expanded stylistic range and prompt adherence. OpenAI’s system card describes stronger steerability and an emphasis on following user direction with high fidelity.

For creators, that meant prompts could describe not only what should appear, but also aspects of how a scene should behave.

The Storyboard Approach Gave Creators More Control

One of Sora’s more interesting interface ideas was Storyboard.

Instead of describing an entire video in a single prompt, the storyboard interface allowed users to specify inputs at different points in the sequence. OpenAI’s product announcement described it as a way to precisely specify inputs for each frame or stage.

This is useful because long prompts can become difficult to manage.

Consider a short narrative with three parts: a character enters a room, discovers something unexpected, and then reacts. A single prompt has to communicate all of those events and their order.

Storyboard lets the creator think more like a director.

The scene can be divided into moments. Each moment can introduce or reinforce the desired visual information. That makes temporal structure easier to express.

The concept also reflects a broader issue in AI video generation: consistency.

Generating one beautiful frame is relatively easy compared with maintaining the same character, location, objects, and visual logic through multiple scenes.

A storyboard does not completely solve consistency, but it gives the creator more explicit control over the structure.

For filmmakers, animators, and visual storytellers, that type of control is more useful than simply generating random clips from one-line prompts.

Remix, Blend, and Loop Made Sora More Experimental

Sora’s creative tools were also designed around transformation rather than constant generation from zero.

Remix allowed users to take an existing generated scene and create another version based on new instructions. Blend could combine two pieces of video into one result, while Loop was designed around creating a repeating sequence.

These features changed the creative process.

Instead of treating every generation as disposable, creators could use one result as the starting point for another.

That is closer to how traditional visual editing works.

A filmmaker rarely throws away a scene after making one variation. They adjust the shot, change the framing, try another take, or experiment with a different visual treatment.

Generative tools can benefit from the same iterative approach.

Remix encouraged that behavior. A creator could begin with an acceptable result, change the setting, alter the action, or request a different style.

Blend provided another kind of experimentation. Rather than choosing one generated sequence, users could combine elements from two clips and explore how the model interpreted the relationship.

Loop was more focused. Repeating movement can be useful for backgrounds, atmospheric scenes, social posts, and short visual assets that do not require a conventional beginning and ending.

Together, these tools positioned Sora as a creative playground as much as a generator.

Sora 2 Added Characters and Consent-Based Likeness

Sora 2 introduced one of its most distinctive product features through “characters.”

OpenAI described characters as a consent-based system that could place a user’s verified likeness into generated scenes. The process involved a short video-and-audio recording in the app to verify identity and capture the person’s likeness. Users could control who was allowed to use their character and could revoke access.

This was a significant product direction because realistic likeness generation raises obvious questions about consent.

Without safeguards, a system capable of generating realistic people can also be used to create misleading or non-consensual material.

OpenAI’s Sora 2 safety documentation therefore emphasized consent, identity protection, provenance, and restrictions around sensitive content. The company also described safeguards around photorealistic people, image uploads, and content involving minors.

The character approach was designed to put more control in the hands of the person whose likeness was being used.

It also introduced an interesting creative possibility.

A user could appear as a character in a fictional environment without filming that environment themselves.

That made Sora 2 useful for playful storytelling, social content, creative skits, and experimental scenes.

At the same time, likeness technology requires stronger safeguards than ordinary fictional character generation. OpenAI’s deployment materials recognized risks involving misleading generations and unauthorized use of someone’s identity.

Sora’s Video Quality Was Closely Tied to Physical Consistency

Sora’s public reputation was strongly connected to how it handled movement.

AI video generation has historically struggled with physical relationships. Objects can change shape. Hands can behave strangely. People can interact incorrectly with furniture or other objects. Motion can become inconsistent between frames.

OpenAI positioned Sora 2 specifically as an improvement in physical accuracy.

The system card describes more accurate physics, sharper realism, better steerability, and stronger handling of dynamic scenes.

This is important because realism in video is not just about resolution.

A high-resolution clip can still look artificial if an object suddenly changes size or a person interacts with something in an impossible way.

Viewers notice those mistakes quickly because video gives the brain many frames of information about motion.

Sora’s approach attempted to address that deeper problem.

However, the system was not perfect.

Independent reviews published after the shutdown still noted limitations with complex actions and human anatomy, while praising areas such as physical simulation, synchronized audio, storyboarding, and character insertion.

That is a useful distinction.

Sora represented significant progress, but progress did not mean every generated sequence became production-ready.

Creators still needed to inspect outputs carefully, especially when scenes involved complicated interactions, hands, multiple characters, or detailed physical actions.

Audio Turned Sora 2 Into a More Complete Generation System

The addition of synchronized sound deserves separate attention because it changed what a generated clip could communicate.

A visual sequence with no sound often feels incomplete. If two characters are having a conversation, the absence of speech makes the result feel more like a preview than a finished scene.

Sora 2 was designed to generate dialogue and sound effects in sync with the generated footage. OpenAI explicitly described synchronized dialogue and sound effects as a key improvement.

This expanded the number of use cases.

A creator could imagine a short dialogue scene, a character reacting to an event, an atmospheric environment, or a fictional sequence with natural sound.

Audio also adds temporal information.

A sound occurring at the right moment reinforces what happens visually. A door opening, footsteps approaching, or a voice responding to an action can make a scene feel more connected.

Still, synchronized audio does not remove the need for editing.

Generated dialogue may need review. Timing may not match the creator’s preferred pacing. Sound design can benefit from additional mixing.

The value is that the initial generation contains both picture and sound, giving the creator a more complete starting point.

Sora Was Designed for Creative Exploration, Not Traditional Editing Alone

One way to understand Sora is to compare it with conventional video software.

A traditional editor starts with media.

Sora started with an idea.

The creator described something that did not necessarily exist yet, and the model attempted to produce it.

That makes it useful for concept development.

A filmmaker could visualize a scene before production. A designer could explore motion ideas. An animator could experiment with a visual style. A content creator could produce a fictional environment without filming it physically.

At the public-product stage, OpenAI also built social discovery into Sora. Sora 2 launched with a standalone iOS app where users could discover, remix, and share creations through a customizable feed.

This gave the product a social dimension that went beyond standard creative software.

Users were not simply generating files.

They were also seeing what other people created and using those creations as starting points.

That model encouraged iteration and remixing, although it also introduced additional questions around moderation, likeness, provenance, and content ownership.

OpenAI’s safety approach included visible and invisible provenance signals. The company said Sora videos included C2PA metadata and other mechanisms designed to help identify generated content.

Pricing Changed Across Sora’s Different Stages

Sora’s pricing cannot be described with a single number because the product evolved through several access models.

When the standalone Sora product launched in December 2024, OpenAI said it was included with ChatGPT Plus and Pro. The public Sora page listed Plus at $20 per month with video generation up to 720p and 10 seconds, while Pro was listed at $200 per month with faster generations, up to 1080p and 20-second videos, concurrent generations, and watermark-free downloads.

Sora 2 later launched under a different access model.

OpenAI said Sora 2 would initially be available for free with generous limits, subject to compute constraints, while ChatGPT Pro users would receive access to the experimental higher-quality Sora 2 Pro model.

Those historical differences matter because many pages on the web still mix the original Sora pricing with Sora 2 access.

They should not be treated as one permanent pricing structure.

Today, however, neither model is available through the former Sora consumer product because OpenAI discontinued the service.

The Most Important Sora Question Today Is Availability

For current users, this is the first thing to check.

Sora’s web and app experiences were discontinued on April 26, 2026. OpenAI also announced the shutdown of the Sora API on September 24, 2026.

That means a user searching for “Sora AI” today should not be given an outdated guide telling them to sign up and begin generating videos.

The product is no longer available as it was.

OpenAI’s discontinuation guidance also explains what happens to existing content. Users were given an export path through the Sora sunset page, and OpenAI states that Sora-related data would be permanently deleted after the applicable final export period.

This changes how the product should be evaluated.

Instead of asking only what Sora can do, a current reader may want to know why it disappeared, what functionality made it notable, and what can be learned from the system.

That makes Sora an important historical case study in generative video.

What Sora Leaves Behind

Sora’s biggest contribution was not one particular interface feature.

It helped move expectations around AI video.

The original system showed that models could produce longer, coherent visual sequences from text. Sora 2 pushed the idea further with synchronized audio, stronger physical behavior, greater steerability, and character-based creation.

It also demonstrated that a video-generation product can be more than a prompt box.

Storyboard, Remix, Blend, Loop, Characters, social discovery, and editing-oriented controls all pointed toward a more interactive form of generative media creation.

At the same time, Sora exposed the limits and risks of the technology.

Realistic video requires careful treatment of likeness, provenance, misleading content, and moderation. OpenAI’s Sora 2 safety materials specifically discussed those challenges and the safeguards designed around them.

For creators looking at Sora today, the most accurate view is therefore historical rather than promotional.

Sora was an important AI video platform, Sora 2 expanded the concept with audio and social creation, and the consumer product is now discontinued.

Its legacy is likely to be measured less by the service that existed for a limited time and more by the expectations it helped establish for what generative video models could eventually produce.