During the National Day holiday， Sora 2 ignited the entire AI community， even being dubbed “AI version of Douyin” and “the next super app” by netizens.
On September 30th， OpenAI released Sora 2， its most advanced video generation model to date， and simultaneously launched an iPhone app called “Sora by OpenAI.” Within a few days， Sora topped the charts on the US App Store， surpassing both OpenAI’s ChatGPT and Google’s Gemini.
Furthermore， with the launch of Sora’s “Cameo” feature， OpenAI CEO Sam Altman became one of the busiest people on the internet. Driven by user creativity， he was seen singing in the toilet， dancing in front of the pyramids， appearing to make pre-packaged meals， and even endorsing Mixue Bingcheng. These hyper-realistic short videos， indistinguishable from reality， went viral across the internet， leading netizens to exclaim they had “played it to death.”
As Sora 2 rapidly gained popularity， AI DeepFakes once again swept the web. Beyond Altman， netizens with wild imaginations seamlessly embedded the animated Detective Conan into the plot of “Prince You Get Mail，” “resurrected” Kobe Bryant for dinner with Pikachu， and even depicted Zhuzhuxia being stopped by traffic police while driving a Xiaomi YU7.
However， behind this frenzy， copyright disputes and the risks of deepfakes have emerged like hidden reefs. Lawyers have warned， “Although the videos are AI-generated， the original IP characters they are based on are themselves protected by copyright. If users engage in commercial activities such as gaining followers or profiting from such videos， it can easily constitute copyright infringement.”
Sora 2 Tops US Charts， 15-Second Blockbusters “Play” with Altman
This week， OpenAI， valued at $500 billion， launched its standalone social app， Sora App， which quickly became a sensation in the global AI community upon its release.
As a standalone mobile application， the Sora App’s core is built upon OpenAI’s latest video generation model， Sora 2， designed to create immersive short video content for users. This means users can generate ultra-realistic short videos up to 15 seconds long by simply uploading a photo or providing a text prompt， effectively making anyone a potential “movie director.”
OpenAI claims that the Sora 2 model features significant upgrades compared to its predecessor: it achieves synchronized audio and visual generation for the first time， greatly enhances physical accuracy， and improves image resolution and detail realism， marking a “GPT-3.5 moment” in AI video generation.
Sora 2 is capable of accomplishing tasks that were exceptionally difficult， or even impossible， for previous video generation models. For instance， it can accurately simulate Olympic gymnastics moves， perform backflips on a paddleboard with precise physics simulation of buoyancy and rigid body dynamics， and render a triple-and-a-half jump by a cat struggling to maintain its grip. For example， a skateboarder leaping into the air can be accompanied by the sound of the wind， the impact， and even the ambient urban noise， all seamlessly integrated.
Moreover， the Sora App is not just an AI tool but also a video-sharing platform that combines social entertainment. Users can create AI-generated audio and video content， scroll through a customizable feed， and “insert” themselves or their friends into videos using the “Cameo” feature.
This technological breakthrough has instantly ignited the creative passion of ordinary users. Leveraging Sora’s newly launched “Cameo” feature， OpenAI CEO Sam Altman has become the busiest person on the internet.
His image has been “manipulated” by netizens in every imaginable way: one moment he’s giving a formal speech in a boardroom， the next he’s appearing in Egypt to dance， and soon after he’s dueling Elon Musk in a virtual battlefield.
On social media， Chinese netizens are engaging in a highly localized creative carnival using Sora. Altman is not only performing Sichuan Opera face-changing， riding an electric bike as a Meituan delivery driver， becoming a beauty blogger， but also enthusiastically advertising for Mixue Bingcheng， and even joining netizens in a popular trending dance called “Skillful Go.”
Data shows that after the launch of Sora’s “Cameo” feature， related UGC content increased by 1200% within 24 hours， with users generating an average of 3.2 videos per day， a frequency far exceeding that of early ChatGPT text interactions.
Currently， Sora 2 is only available to invited users. The iOS app can be downloaded， and the web version can be accessed on Android devices. Initial rollout will be in the US and Canada， with expansion to more countries planned.
It is worth noting that OpenAI’s recent financial disclosures revealed revenues of $4.3 billion in the first half of the year， exceeding the entire previous year’s total， but also a net loss of $13.5 billion. OpenAI CFO Sarah Friar has publicly stated that the company is considering an IPO at some point in the future. OpenAI is currently facilitating a secondary share sale for its employees， with the company’s valuation reaching $500 billion.
Lawyers Warn: High Risk of Copyright Infringement
However， following the launch of the Sora App， many videos have also raised concerns among copyright and deepfake experts.
In addition to Altman， globally recognized IPs like Pikachu have become popular subjects for Sora 2 creations， with netizens portraying Pikachu as everything from a soldier to a chef and a pirate. Others have integrated video game characters like Lara Croft， or Nintendo heavyweights such as Mario， Luigi， and Princess Peach， into their Sora 2 creations.
Furthermore， netizens with exceptional creativity have seamlessly embedded the animated Detective Conan into the plot of “Prince You Get Mail，” had the Teletubbies enjoy lamb skewers on Prairie Home， and even depicted Zhuzhuxia being stopped by traffic police while driving a Xiaomi YU7.
The proliferation of derivative IP content on the Sora platform is essentially a product of the collision between OpenAI’s “opt-out” mechanism and copyright protection rules. On the Sora App， an “opt-out” rule is applied to copyrighted content: the platform assumes all copyrighted characters can be generated by default， and only copyright holders can prevent their IPs from being used by actively contacting OpenAI to request their removal.
Guo Qinggui， a renowned lawyer and finance writer， stated to the “BUG” column， “This carries a high risk of copyright infringement， and both the platform and individual users must proceed with caution.”
Guo Qinggui elaborated， “If Sora is used to generate videos containing copyrighted characters (such as Pikachu)， authorization must be obtained， otherwise it may constitute infringement. According to Article 10 of the Copyright Law， copyright holders enjoy rights such as reproduction， adaptation， and communication through information networks. When users generate videos with copyrighted characters (like Pikachu) using Sora without the copyright holder’s authorization， they may infringe upon the rights of reproduction (recreating the character’s image)， adaptation (recombining the character with scenes)， and communication through information networks (uploading to the platform or other channels).”
If the video includes real individuals’ images， such as the user themselves or passersby， attention must also be paid to Article 1019 of the Civil Code concerning the right of portrait: using someone’s likeness without consent may constitute infringement. He mentioned， “Even if AI technology is used to alter facial features， if the identity is still recognizable， authorization is still required. In general， user behavior on the Sora App that merges real-life images with AI-generated scenes may infringe upon copyright， portrait rights， and privacy rights if copyrighted characters or other people’s likenesses are involved.”
Furthermore， OpenAI’s “opt-out” copyright strategy may not truly circumvent legal risks.
Guo Qinggui noted that the “opt-out” rule adopted by the Sora App cannot entirely avoid legal risks. Although this model aligns with the “safe harbor principle” framework in some regions， it still presents multiple hidden dangers in judicial practice， copyright holder enforcement strategies， and international regulatory trends.
According to the US Digital Millennium Copyright Act (DMCA) and China’s “Regulations on the Protection of the Right of Communication Through Information Networks，” platforms can be exempt from liability for damages if they merely provide technical services and fulfill the “notice and takedown” obligation. However， Sora’s “opt-out” mechanism is essentially “infringe first， deal with it later.” This means it allows users to generate copyrighted content by default and only blocks IPs after copyright holders actively request it， a model that fundamentally differs from “notice and takedown.”
He provided a specific example: OpenAI has argued that the use of AI training data constitutes “fair use.” However， this defense faces challenges in judicial practice: first， the lack of transformative use. Videos generated by users， such as “Pikachu fighting Ultraman，” are typically for entertainment and do not involve critical or educational modification of the original work， making it difficult to meet the requirements of “fair use” under US copyright law. Second， there is a commercial profit-making aspect. Sora’s user agreement indicates that the platform may profit from infringing content through advertising revenue sharing， subscription fees， and other means. Such commercial use weakens the basis for a “fair use” defense.
“In the context of copyright holders actively defending their rights， tightening international regulations， and judicial practices leaning towards protecting rights holders， platforms still face multiple pressures from lawsuits， administrative penalties， and loss of commercial reputation. OpenAI needs to make systematic adjustments on technical， legal， and business model levels， finding a balance between innovation and compliance through proactive filtering， tiered authorization， and originality incentives. Otherwise， Sora may repeat the fate of Midjourney and get bogged down in copyright disputes，” lawyer Guo Qinggui stated.
