Sora

Sora was OpenAI’s branded video-generation product, introduced publicly in February 2024 as a model capable of converting written descriptions into realistic moving images. It was designed to generate short video clips from natural-language prompts, extend existing footage, and support creative experimentation involving camera movement, visual style, setting, and action. The name was derived from the Japanese word for “sky,” reflecting OpenAI’s stated association with expansive creative potential. The product emerged from OpenAI’s broader generative-AI research program, following the company’s development of DALL·E image-generation systems. Its technical approach adapted diffusion-model methods to video and used a transformer architecture to process compressed spatiotemporal representations. Rather than generating every video frame independently, the system operated on three-dimensional patches representing visual information across space and time. OpenAI also described the use of detailed video captions to improve the relationship between prompts and generated scenes. OpenAI previewed Sora on February 15, 2024, releasing examples that included cinematic street scenes, historical-style footage, animated creatures, vehicles, and people moving through snow-covered environments. The demonstrations attracted attention because they showed stronger temporal consistency, camera behavior, and prompt adherence than many earlier text-to-video systems. At the same time, OpenAI acknowledged that the model could struggle with complex physical interactions, causality, object permanence, and distinctions such as left versus right. The previews also raised concerns about synthetic misinformation, employment in film and advertising, copyright, and the impersonation of real people. Before public availability, OpenAI provided restricted access to safety researchers, misinformation and bias experts, and selected creative professionals. The purpose was to stress-test the system and collect feedback from artists, filmmakers, and other potential users. Sora’s first public release began on December 9, 2024, for ChatGPT Plus and ChatGPT Pro customers in the United States and Canada. The product subsequently developed into a broader user-facing experience rather than remaining only a research demonstration. OpenAI announced plans in February 2025 to connect Sora more closely with ChatGPT-based creation workflows. Sora 2 was introduced on September 30, 2025. It combined upgraded video generation with audio-related capabilities, character or “cameo” features, and a social feed built around sharing generated clips. The accompanying mobile application was positioned as a social-media-style creative product, with commentators comparing its browsing and sharing mechanics to TikTok and other short-video platforms. Sora 2 expanded the product’s reach and cultural visibility, but also intensified disputes over copyright, deepfakes, likeness rights, watermark circumvention, and the possible use of synthetic video in scams or political disinformation. The Sora web and application experiences were discontinued on April 26, 2026. OpenAI scheduled the Sora API for discontinuation on September 24, 2026, meaning that the brand remained in a transitional state during the final API period but no longer operated as a consumer web or mobile service. OpenAI did not give a detailed public explanation for the shutdown. Contemporary reporting connected the decision with the high computational cost of video generation, constrained access to computing resources, declining usage after the initial viral period, and a strategic preference for core enterprise products. Sora therefore became a notable but short-lived example of a major technology company attempting to turn frontier video generation into a mass-market social product.

Sora developed from OpenAI’s effort to extend generative modeling beyond still images into moving, spatially coherent scenes. Its immediate technical lineage was connected to DALL·E 3, but the product addressed a more difficult problem: generating a sequence of images that remains visually consistent while also representing motion, objects, camera changes, and interactions over time. OpenAI described the resulting system as a diffusion transformer. In simplified terms, it generated compressed three-dimensional patches representing space and time, progressively denoised those representations, and then decoded them into ordinary video. The product was publicly introduced on February 15, 2024. OpenAI released a collection of demonstration clips showing urban scenes, animals, vehicles, animation, historical reconstructions, and cinematic compositions. The company said Sora could produce videos up to one minute long while maintaining visual quality and adherence to the prompt. The demonstrations made Sora one of the most visible text-to-video systems of its period, although reviewers cautioned that showcase examples were likely selected for quality and did not necessarily represent typical output. OpenAI also published a technical report describing the model’s training and architecture. During its pre-release phase, Sora was made available to a small group of red-team participants, including specialists in misinformation, bias, and harmful content. OpenAI also worked with a limited number of artists, filmmakers, and other creative professionals. This program was intended to identify misuse risks and assess whether the tool could support real creative work rather than merely produce impressive demonstrations. The arrangement became controversial in November 2024 when a group of testers leaked an access key and published a protest manifesto accusing OpenAI of using artists to legitimize or improve the system without sufficient recognition of their concerns. OpenAI revoked the exposed access and said participation in the program was voluntary. On December 9, 2024, OpenAI began public distribution of the first-generation product to ChatGPT Plus and ChatGPT Pro users in the United States and Canada. Sora was presented as a standalone video-generation experience, while OpenAI also discussed ways to connect it with ChatGPT. Users could create clips from text and, depending on the version and interface, work from images or existing videos and extend footage forward or backward. The service was aimed at a wide range of use cases, including storyboarding, concept development, advertising mockups, social content, visual experimentation, and previsualization. Sora’s capabilities were accompanied by visible limitations. OpenAI acknowledged difficulty with complex physical reasoning, causal relationships, object interactions, and spatial distinctions. The system could create highly convincing surfaces and camera movements while still producing errors in anatomy, object continuity, or the logic of an action. These limitations mattered because the product’s realism made errors potentially misleading. OpenAI therefore used safety filters, provenance signals, C2PA metadata, and, in many cases, visible moving watermarks. The company also restricted or moderated prompts involving sexual, violent, hateful, celebrity, and existing intellectual-property content, although the effectiveness and consistency of those safeguards became subjects of public debate. Sora 2 was launched on September 30, 2025. It represented a shift from a primarily creative model interface toward a consumer platform with audio, social discovery, remixing, and character-related features. The first Sora 2 application appeared on iOS, followed by an Android application approximately two months later. Its feed-based design led many observers to compare it to TikTok and other short-video networks. Supporters saw the social interface as a way to make advanced video creation accessible to people without traditional production skills. Critics described the service as a channel for low-quality synthetic media, deepfakes, misinformation, scams, and unauthorized reuse of recognizable characters. Copyright became one of the product’s most persistent disputes. OpenAI’s training disclosures referred to publicly available videos and copyrighted material licensed for training, without publishing a complete list of sources. In the Sora 2 product, copyrighted characters and recognizable fictional properties could become part of generated videos unless rights holders sought restrictions. OpenAI later said it would provide more granular controls, but the policy drew criticism from entertainment-industry representatives and rights organizations. In Japan, a content-distribution association reportedly requested that OpenAI stop using works associated with member companies, including major animation and game properties. Separate controversies involved videos depicting the likenesses of deceased celebrities, prompting families and estates to request stronger protections. The product also became part of a broader labor and cultural debate. Filmmakers and entertainment workers argued that generative video could reduce demand for writers, artists, editors, designers, performers, and production staff. Some commentators emphasized the opposite possibility: that lower technical barriers could allow more people to express ideas and participate in visual storytelling. The debate was sharpened by public criticism from creators, including claims that AI video could accelerate the spread of “slop” content and weaken professional standards. Concerns about deepfake political material were particularly significant because realistic video can be used to imitate public figures or create fabricated events. OpenAI announced in March 2026 that it would discontinue Sora. The web and application experiences ended on April 26, 2026, while the API was scheduled to remain available until September 24, 2026. The shutdown notice did not provide a detailed product-level explanation. Reporting associated the decision with compute scarcity, operating expense, declining engagement after the initial launch, and a broader shift toward enterprise-oriented products. The closure also affected the expected future of the Disney licensing relationship. Sora consequently ended as a consumer social product less than two years after its first public release, while its research ideas and generated-media safety practices remained part of the wider generative-AI landscape.

• 2026

  Sora web and app experiences shut down

  [4]What to know about the Sora discontinuationhelp.openai.com

  OpenAI discontinues the Sora web and application experiences on April 26, 2026.

• 2026

  Sora API scheduled for discontinuation

  [4]What to know about the Sora discontinuationhelp.openai.com

  OpenAI schedules the Sora API for shutdown on September 24, 2026, completing the planned withdrawal of the Sora product line.

• 2025

  OpenAI announces deeper ChatGPT integration plans

  [5]Sora (text-to-video model)en.wikipedia.org

  OpenAI announces plans to make Sora generation more closely available through ChatGPT-based creation workflows.

• 2025

  Sora 2 launches with a social application

  [2]Sora 2 is hereopenai.com

  Sora 2 is unveiled on September 30, 2025, alongside an iOS application that combines video generation with sharing, remixing, audio, and character-related features.

• 2025

  Android application follows the iOS launch

  [5]Sora (text-to-video model)en.wikipedia.org

  An Android version of the Sora application is released approximately two months after the Sora 2 launch.

• 2024

  OpenAI publicly previews Sora

  [1]Sora: Creating video from textopenai.com

  OpenAI introduces Sora on February 15, 2024, with demonstrations of high-definition videos generated from text prompts and a technical report describing the model.

• 2024

  Pre-release safety and creative testing

  [1]Sora: Creating video from textopenai.com

  OpenAI gives restricted access to red-team researchers and selected creative professionals to examine misinformation, bias, harmful content, and practical usefulness.

• 2024

  First-generation Sora launches publicly

  [1]Sora: Creating video from textopenai.com

  On December 9, 2024, Sora begins rolling out to ChatGPT Plus and ChatGPT Pro users in the United States and Canada.

A frontier generative-AI video platform positioned between a creative-production tool, a research model for world simulation, and a consumer social-video application.

• Text-to-video generative AI model · 2024

  Sora

  The first publicly demonstrated Sora model converted natural-language descriptions into short video clips. It was designed to preserve visual quality and prompt alignment across time, and could depict cinematic scenes, animated subjects, environments, vehicles, and human activity. The model also supported extending or continuing existing short videos in certain workflows. OpenAI presented it as both a practical creative tool and a research step toward systems that understand and simulate aspects of the physical world. Its known limitations included weak handling of complex physics, causality, object interactions, and some spatial relationships.

• Multimodal video-generation model and social media platform · 2025

  Sora 2

  Sora 2 was the second major generation of OpenAI’s video system, launched on September 30, 2025. It extended the product toward a consumer social experience by adding a dedicated mobile application, audio-related generation, sharing, remixing, and cameo or character functions. The service used visible and invisible provenance measures, including moving watermarks on many videos and C2PA metadata. Its public reception combined praise for accessibility and realism with criticism concerning copyright, deepfakes, low-quality synthetic content, and the resemblance of its feed to established short-video platforms.

• Consumer social video application · 2025

  Sora mobile application

  The Sora application turned video generation into a feed-based consumer product. Users could create, view, share, and remix synthetic videos, with the service emphasizing social discovery and participation rather than only professional production. The application was initially released for iOS with Sora 2 and later expanded to Android. Its social design broadened the audience for generative video while increasing concerns about content moderation, impersonation, copyright enforcement, and the rapid circulation of fabricated audiovisual material.

• Developer API

  Sora API

  The Sora API provided programmatic access to OpenAI’s video-generation capabilities for developers and product builders. It represented the infrastructure-oriented side of the brand, allowing Sora functionality to be incorporated into external applications and automated workflows. OpenAI announced that the API would be discontinued on September 24, 2026, after the consumer web and application experiences had already ended.

• 2025

  Copyrighted-character generation in Sora 2

  Sora 2 generated videos involving recognizable copyrighted characters and properties. Rights holders and industry representatives criticized the default treatment of protected material and called for stronger opt-out and authorization controls.

  Source

• 2025

  Deepfake likenesses and deceased celebrities

  Families and estates associated with deceased celebrities objected to hurtful or unauthorized synthetic videos depicting their relatives. The controversy led OpenAI to restrict some figures and provide additional mechanisms for estates to request protections.

  Source

• 2025

  Watermark circumvention

  Third-party programs reportedly emerged soon after the Sora 2 release that could remove or obscure visible moving watermarks, raising concerns about provenance, attribution, and the misuse of generated videos.

  Source

• 2025

  Creative-industry employment concerns

  Film professionals, performers, writers, artists, and labor organizations warned that high-quality AI video could reduce demand for human production work. Other commentators argued that the technology could lower barriers to entry and expand participation in visual storytelling.

  Source

• 2024

  Tester access-key leak and artist protest

  A group of testers publicly disclosed an access key and criticized OpenAI’s use of artist participation in Sora’s development. OpenAI revoked the compromised access and defended the testing program as voluntary.

  Source

• 2024

  Training-data and copyright controversy

  OpenAI said Sora was trained using publicly available videos and copyrighted videos licensed for training, but did not publish a complete account of the data sources. The lack of detailed disclosure contributed to wider disputes over consent, compensation, and the use of creative works in generative-AI training.

  Source

Read more

• 2025Sora 2 social-video launch

  United States · Canada

  Sora 2 launched with a mobile-first social experience designed to encourage users to generate, browse, share, and remix short AI videos. The product emphasized participation and network effects in addition to model quality.

  Outcome. The campaign expanded Sora’s cultural visibility and drove a viral user response, but it also intensified criticism that the product encouraged AI-generated “slop,” deepfakes, and unauthorized use of fictional or real-person likenesses.

  Source
• 2024Sora public preview

  Global media audience

  OpenAI introduced Sora through a curated set of cinematic demonstration videos, including urban scenes, vehicles, animation, historical-style footage, and human movement. The launch functioned as both a research announcement and a brand-building campaign for OpenAI’s entry into high-fidelity video generation.

  Outcome. The preview generated extensive international attention and established Sora as one of the best-known early text-to-video brands, while also prompting debate about misinformation, creative labor, and copyright.

  Source
• 2024Creative professional and red-team access program

  United States · Canada · Selected international participants

  OpenAI invited selected artists, filmmakers, and safety specialists to test Sora before broad public release. The program sought feedback on creative usefulness and adversarial testing of misinformation, bias, and harmful-content risks.

  Outcome. The program informed safety work and public discussion, but some participants later protested what they viewed as exploitative or symbolic artist involvement, culminating in the November 2024 access-key leak.

  Source

• 2026

  Discontinue Sora web, app, and API productsStrategy

  Background

  OpenAI announced the end of Sora after the product’s consumer phase, without providing a detailed public explanation in the discontinuation notice.

  Action taken

  The Sora web and application experiences ended on April 26, 2026, and the Sora API was scheduled to end on September 24, 2026.

  Aftermath

  The decision ended Sora as an active consumer brand and was associated in reporting with compute constraints, high operating costs, declining usage, and a shift toward enterprise priorities.

  Reported estimated operating cost. Approximately $1 million per day (Reported in connection with the 2026 discontinuation; estimate, not an OpenAI-reported audited figure)

  Source

• 2025

  Shift from model interface to social video platformGeneration change

  Background

  OpenAI sought to broaden Sora’s audience and make generative video a participatory consumer activity rather than only a specialist creative workflow.

  Action taken

  The company launched Sora 2 with a dedicated social application, sharing feed, remixing functions, audio-related capabilities, and character or cameo features.

  Aftermath

  The move increased visibility and user participation but exposed the product to intensified moderation, copyright, deepfake, and platform-safety challenges.

  Source

• 2025

  Introduce more granular copyright controlsStrategy

  Background

  Sora 2’s ability to generate recognizable copyrighted content drew criticism from rights holders and entertainment-industry organizations.

  Action taken

  OpenAI said it planned an update that would give copyright holders more detailed control over how their content could be generated, while the precise treatment of existing material remained unclear.

  Aftermath

  The announcement acknowledged the scale of the copyright dispute but did not resolve concerns about default permissions, training data, or the boundary between parody, transformation, and infringement.

  Source

• 2025

  Enter a character-licensing arrangement with DisneyM&A

  Background

  OpenAI pursued licensed intellectual property as a way to make recognizable characters available in a controlled commercial framework.

  Action taken

  Disney announced a three-year licensing arrangement involving more than 200 characters from Disney Animation, Pixar, Marvel, and Star Wars properties, alongside a reported investment in OpenAI.

  Aftermath

  The arrangement was welcomed by some as a move toward permission-based generative media, while labor organizations and creative professionals criticized it as a threat to human work and creative rights.

  Reported Disney investment in OpenAI. $1 billion (Announced December 11, 2025)

  • Walt Disney Company — Announced the licensing and investment arrangement involving OpenAI and Sora-related character generation.

  Source

• 2024

  Preview Sora before public releaseProduct launch

  Background

  OpenAI wanted to demonstrate progress in video generation while allowing external safety researchers and selected creators to evaluate the system before wider deployment.

  Action taken

  The company released curated examples and a technical report but initially limited direct access to selected testers.

  Aftermath

  The strategy created substantial public interest and gave OpenAI time to study misuse risks, but the curated nature of the examples also led reviewers to question how representative they were.

  Source

• 2024

  Release Sora to ChatGPT Plus and Pro customersProduct launch

  Background

  After restricted safety and creative testing, OpenAI moved Sora from a research preview toward a paid consumer product.

  Action taken

  Beginning December 9, 2024, OpenAI rolled out Sora to eligible ChatGPT Plus and ChatGPT Pro users in the United States and Canada.

  Aftermath

  The release made high-end text-to-video generation available to a larger audience and positioned Sora within OpenAI’s commercial product portfolio.

  Source

• 2026

  OpenAI discontinues Sora web and app experiences[4]What to know about the Sora discontinuationhelp.openai.com

  Other

  OpenAI ended the Sora web and mobile experiences on April 26, 2026, and announced that the Sora API would be discontinued on September 24, 2026.

• 2025

  OpenAI announces Sora 2 and a social video application[2]Sora 2 is hereopenai.com

  Product launch · Product generation

  On September 30, 2025, OpenAI introduced Sora 2 alongside a new iOS application focused on generated video, sharing, remixing, and user-created character appearances.

• 2025

  Disney announces investment and licensing arrangement involving OpenAI[5]Sora (text-to-video model)en.wikipedia.org

  M&A · Other

  The Walt Disney Company announced a three-year licensing arrangement involving OpenAI that would permit the use of more than 200 characters from Disney, Pixar, Marvel, and Star Wars properties in Sora-related generation workflows.

• 2024

  OpenAI previews Sora text-to-video model[1]Sora: Creating video from textopenai.com

  Product launch · Product generation

  OpenAI publicly demonstrated Sora-generated high-definition clips and described a system capable of producing videos of up to approximately one minute from written prompts.

• 2024

  Sora access expands to selected creative professionals and safety testers[1]Sora: Creating video from textopenai.com

  Other

  Before general release, OpenAI gave limited access to red-team researchers and creative professionals to evaluate safety, misinformation risks, bias, and usefulness in artistic workflows.

• 2024

  Sora becomes available to ChatGPT Plus and Pro users in the United States and Canada[1]Sora: Creating video from textopenai.com

  Product launch

  OpenAI began its first public rollout of Sora on December 9, 2024, making the service available to eligible ChatGPT Plus and ChatGPT Pro users in the United States and Canada.

1. 1Sora: Creating video from textopenai.com
2. 2Sora 2 is hereopenai.com
3. 3Creating with Sora safelyopenai.com
4. 4What to know about the Sora discontinuationhelp.openai.com
5. 5Sora (text-to-video model)en.wikipedia.org
6. 6Video generation models as world simulatorsopenai.com
7. 7Sora Secures $2.7M in Funding to Set Gold Standard in Virtual High School Educationprweb.com