OpenAI is not a gray company. When they presented the first version of Sora, it seemed that this video AI, associated with ChatGPT, was going to change everything. However, the reduced model that eventually arrived 10 months later was a complete fiasco. Now, they have launched Sora 2 without prior notice, and I can’t stop using it. It is totally addictive, with a magical potential that far exceeds the launch of Google Nano-Banana . It reminds me of the Studio Ghibli phenomenon associated with ChatGPT, making me think that this time, OpenAI had everything calculated.
You Must Make an Effort to Access Sora 2
In the absence of an API to access the model through third-party tools, the only way to create videos with Sora 2 is through Sora, a new OpenAI social network functioning like TikTok, but with AI-generated videos. However, Sora is currently available only in the United States and Canada . To access it, I had to install a VPN. After selecting a username (or importing the one from the old Sora), uploading a profile picture, and crafting a bio, I could finally start generating videos with nine seconds of sound in either panoramic or vertical format .
However, not so fast. Sora’s web version lacks several functions, including the most popular: creating a “cameo” with your face and voice to become the protagonist in your own videos or appear in the videos of others. To configure that option, you’ll need to download the Sora app, which is currently exclusive for iPhone and only available in the United States and Canada App Store. Therefore, I had to install a VPN on my iOS device, create an Apple account in the United States, associate it with the App Store, and download Sora.
Once inside, I found a familiar interface: a Feed of recommended content, a chronological feed of publications, and the ability to follow specific users. Besides creating your own content, you can respond to comments, etc. Notably, there’s a new feature that allows you to inform the AI about your mood , influencing the type of content you see.
OpenAI Hits the Nail with the Cameos
Anyone who has generated images and videos with AI in recent years knows that models often change your face and struggle to maintain consistency across different scenes. Recently, Loras have become a solution by training a model with your photos (technically refining an already trained model) to generate videos featuring your face. However, Loras can be complex and costly, prompting the emergence of new techniques to improve model consistency. Solutions like Flux Kontext from Black Forest Labs or Gemini 2.5 Flash Image Nano-Banana from Google illustrate this evolution.
Sora’s cameos are a game-changer. Upon opening Sora’s application, you can directly configure your cameo using your phone’s camera by reading numbers displayed on the screen and moving your head as directed. This recording is used to generate videos featuring your face and voice in various contexts. OpenAI has even provided a comprehensive guide to improve your cameos, with the most critical tip being to record in a well-lit room.
With just a single recording, the results can be convincing, but you can also add personalized instructions in a text box—like “I am smiling” or “I am serious.” One crucial aspect is deciding who can use your cameo for video creation: just you, your approved friends, or all users. Notably, Sam Altman, OpenAI’s CEO, has a public cameo that has heavily influenced Sora’s initial success. Many popular videos feature Altman as the main character, including comedic moments depicting him in outrageous situations.
Sora’s introductory video showcased OpenAI engineers overcoming dragons or riding giant ducks. The platform allows users to have multiple cameos, which could enable users to mix their faces with celebrities in various contexts. However, finding celebrities willing to lend their images remains a challenge. Users can produce as many as 100 videos a day, managing them as drafts or publishing them instantly.
More Like Vine than TikTok or Instagram
Sora seems to be an aimed shot at TikTok . Not only has TikTok lured OpenAI engineers with sky-high salaries, but it recently launched its social network for self-generated content, “Vibes.” While Vibes is seen as more artistic, Sora has often been perceived as a collection of AI-generated “garbage.” Yet, using Sora evokes the vibrant nostalgia reminiscent of Vine , an era where short videos held a unique innocence devoid of monetization and competition from countless other feeds.
For now, users are on Sora for fun and to poke fun at Sam Altman. Some users seem to predict that Altman might eventually face legal issues due to his cameo. A series of videos cleverly placed Altman among copyrighted characters , where he humorously comments on the situation, such as hoping that “*Nintendo doesn’t sue us*.”
The app includes features to enable users to replicate others’ content, allowing for video remixes and defaulting the description to the Prompt used. The model appears designed for humor, unlike other AI models that often conclude with surrealist endings. For instance, in one video where I argued with a cashier, the abrupt silence added an unexpected comedic twist.
This AI has Absorbed Just About Everything
Sora’s apparent indifference to copyright constraints is striking. While it doesn’t officially permit the creation of videos featuring the Simpsons or Dragon Ball , users have discovered ways to incorporate these characters into their creations. This has raised eyebrows at OpenAI, leading to potential legal concerns from character producers. OpenAI managed to evade issues with the Studio Ghibli phenomenon since styles are not copyrighted, but characters certainly are.
The creations from Sora encompass an extensive range of topics—everything from fictional scenarios involving political figures to outrageous parodies of pop culture. Users can input a starting image, enhancing complexity. I’ve seen videos showcasing fictional news presenters using real data in a way previously unattainable without considerable resources.
Sora 2: A Major Leap Forward
The latest model, Sora 2, substantially enhances what OpenAI has provided so far. Its multimodal features allow for video and voice generation from text, images, or user cameos. It excels in understanding physical environments better than any existing models. One of the most popular videos features a boy performing a backflip on a surf paddleboard, making it challenging to determine that it’s AI-generated—perhaps noticeable only by scrutinizing the water’s waves.
While Sora 1 promised grand simulations, it often failed to deliver accurate physics, displaying gymnasts attempting impossible contortions. In contrast, Sora 2 is inching closer to realizing its ambitious promises and best of all, it’s accessible to anyone (with a VPN and iPhone).
The mix of personal images and other objects in Sora’s generations is addictively entertaining . It enables the creation of engaging content in mere seconds. Furthermore, OpenAI has adeptly packaged this innovative technology within a social network that, surprisingly, lacks that cringe-factor often overshadowing such platforms. While many videos do suffer from artifacts typical of faster generation processes, many others are genuinely enjoyable.
It’s essential to recognize that for OpenAI, the more critical product isn’t solely Sora 2 (though it showcases impressive technological advancement). It’s about Sora, the social network itself. A successful platform that draws in millions of users and generates advertising revenue could become pivotal for the company behind ChatGPT, proving a direct catalyst towards future monetization.
Can I envision endlessly scrolling through AI-generated content in the days following Sora’s initial surge? If the amusement persists, I can confidently say yes. It’s entertaining, it inspires creativity, and it carries a viral potential reminiscent of Vine’s zenith. I am already creating private memes with friends using our cameos.
Do I contemplate engaging with content that, while original, perhaps wouldn’t exist without the influence of television, YouTube, TikTok, and Reels that OpenAI likely utilized to train Sora 2? It’s possible this article will contribute to shaping future OpenAI models, prompting deeper discussions surrounding ethical concerns inherent in AI content creation. Nevertheless, one thing is clear: the little man with blue eyes, who has overtaken my Feed , knows how to market and launch successful products—especially if it means overshadowing rivals like Mark Zuckerberg along the way.
Images © Sora