Openai is not a gray company. When he presented Sora’s first version, it seemed that the video associated with Chatgpt was going to change everything, but the reduced model that ended up arriving at users 10 months later was a complete fiasco.
Now have launched Sora 2 without prior notice And, well, I can’t stop. It is totally addictive. It has a metic potential that far exceeds The launch of Nano-Banana de Googleand that reminds the Studio Ghibli phenomenon of Chatgpt, which makes me think that, this time, OpenAi had everything coldly calculated.
You have to have effort to get to Sora 2
In the absence of API to get access to the model in third -party tools (Freepik claims to be in it), The only way to create videos with Sora 2 is through Sora, a new OpenAi social network that works like Tiktokbut with videos generated by artificial intelligence itself.
Sora is only available, for the moment, In the United States and Canadaso the first thing I had to do to access is to install a VPN. After choosing a username (or importing the one I already had in the old Sora), uploading a profile picture and putting on a bio, I could finally start generating videos with nine seconds sound, in panoramic or vertical format.
Not so fast. Sora’s web version is missing several functions, including the most popular of all: create a “cameo” with your face and your voice to become the protagonist of your own videos or to appear in the videos of others. To configure that option you have to download the Sora app, which is now exclusive for iPhone, and is only available in the United States and Canada app store. So I had to install a VPN in iOS, create an Apple account in the United States, associate the new account to the app store and download Sora.
Once inside, I found a very familiar interface. A Feed of recommended content, another to see the publications in chronological order and another to see only what you follow. In addition to creating your own content, answering comments, etc. As a novelty, he has an option to tell the AI in what “Mood” you are: what kind of content you want to see.
Openai has given in the nail with the “cameos”
Anyone who has generated images and videos with AI in recent years knows that models tend to change your face. Nor are they very good to maintain the same face in different scenes. Until recently, the solution was the Loras: Train a model with your photos (technically, refine an already trained model) to generate videos with your face.
But Loras are a complex and expensive solution, which has resulted in new techniques to improve the consistency of image models. Flux Kontext of the German company Black Forest Labs or Gemini 2.5 Flash Image Nano-Banana of Google are paradigmatic examples that have pushed the entire sector to improve their consistency. And now Sora cameos arrive.
When you open Sora’s application for the first time, you can directly configure your cameo with the phone camera, reading some numbers that appear on the screen and turning the head in the addresses indicated by the app. The recording is used to generate videos with your face and your voice in the context you can think of. There is A huge guide OpenAi to improve your cameos, but the most important point is to be in a well -lit room.
With the recording enough for the results to be convincing, but you can add personalized instructions in a text box: I am smiling, I am serious, I am fat, I am skinny, etc. The most important thing is to select who will be able to use your cameo to make videos: only you, only the people to whom you have approved access, all your friends or all users.
Sam Altman, the OpenAi CEO, has a public cameo that carries Sora’s initial success. Almost all the videos that have been viralized have Altman as the protagonist. Sora’s first great meme are uncomfortable videos of Sam desperate because they give her more servers: Sam Altman stealing graphics cards in a target, Sam Altman asking for donations, Sam Altman screaming at the Nvidia headquarters that need more GPUS.
He Presentation video Sora 2 included OpenAi engineers flying over dragons or riding giant ducks. The same generation of Sora can have several cameos, so one of the star functions could end up being mixed with celebrities. The problem is to find celebrities willing to give your image that way. In one of my videos, Ijustine tells me How fine the iPhone Air and Sam Altman replies that “it is a respectable size, I would say.” The videos take a few minutes to generate. You can do several at the same time (up to 100 a day) and then publish or leave them in drafts.
More similar to I came than Tiktok or Instagram
Sora is first and foremost an poisoned dart against goal. The queen of social networks has not only been fishing Openai engineers with her astronomical salaries, but just a week ago she had just launched her own social network of self -generated content: “Vibes.” With a more creative or artistic approach that Sora, Vibes has been described by the less flattering comments as a “Ai Slop” landfill, garbage content generated with AI.
Sora is still, in large part, garbage generated with AI. But the feeling I have while I do Scroll It is to have returned to the times of Vine, the precursor of Tiktok that was bought by Twitter and then close. An amalgam of scathing sketches and anonymous moments with the innocence of a time in which the short video did not monetize nor competed for your attention against millions and millions and millions of other vertical videos.
For now, people are in Sora to pass it well. And to put Sam Altman in trouble. Some seem to have bet short for Altman to end up sued and forced to after his cameo. A series of videos shows the OpenAi CEO surrounded by characters protected by copyright: Pokémon, spongebob … and in all videos, “remixes” of the original, Altman ends up saying something like: “I hope Nintendo does not demand us.”
The app has a couple of mechanisms to facilitate the “imitation” of the content of others. In addition to the option to create remixes of a video, the description of a video is by default the Prompt that the user used. Surprises the ease of understanding the intentionality of a Promptwhich can be a few indications.
I have the feeling that the model is tuned so that both the video and the voice have a humorous component. Unlike Vista 3 from Google, which usually ends videos with surrealist situations brand of the house, Sora has abrupt cuts or quite the opposite, which works better. In this video in which I argue with a Mercadona cashierI did not ask Sora that the final silence was so uncomfortable and funny. And in this one in which I ask someone to use their bathroomNor did I specify that there was a side look at the end. He simply did so.
This AI has been seen until the South Canal programs
Sora’s apparent laxity with the copyright is also drawing user attention. The app does not let you generate a video of the Simpsons or Dragon Ball, but many find words that are not censored to end up putting characters such as Super Mario, Peter Griffin, Pikachu or Rick and Morty in surreal situations.
The thing has become uncomfortable for Openai. A Reddit user managed to mix characters from a lot of different anime in the same video, such as Dragon Ball and One-Punch Man, with impressive results that have surely attracted the attention of the producers of those protected audiovisual works. Openai got rid of have problems with the Studio Ghibli phenomenon Because a style is not protected, but the characters are.
This AI has seen all the content you can imagine. From the non-do files (The Generalissimo signs an agreement with Sam Altman to give the water from swamps) to current television (Sam Altman goes to First Dates to look for the AGI), through the South Canal programming with Jesús Quintero and the giggles parting with laughter. I have even achieved that Put my brazacos and tattoos by Jordi Wild. Of course, people’s faces are never those of the real characters; Only with animated characters you can fool the model.
If the Prompt is censored or not enough, another mechanism to add complexity to the videos is the option of loading a starting image. In this video, an news presenter who does not exist makes an informative note From a real graph of intention to vote of the different political parties. Nothing that could not make other models, but with much greater accessibility through Sora. And most importantly: for now, free.
Sora 2 is not as important as Sora
The model, Sora 2, is much better than what Openai offered so far. It is multimodal: it can generate video and voice from a text, an image or your cameo. And understand very well the physical. According to Openai, better than any other model. One of Sora 2’s most watched videos shows a boy making a Backflip On a surf paddle table. It is very difficult to notice that it is made with ia. Maybe if you look a lot at water waves.
Sora 1 had promised us a simulation of the world, but at the moment of truth it made the gymnasts make physically impossible torsions or that the balls disappeared or multiplied when bounted. Sora 2 is much closer to the original promise and, what is more surprising, is available to anyone (at least, with a VPN and an iPhone).
The ease of mixing not only your image, but objects of your photos in Sora’s generations is addictive. It allows to create entertaining or evocative content from an idea in seconds. And for the first time someone has packaged all this in a social network that does not give CRINGE. Yes, many videos are garbage or have artifacts of a quick generation to reduce servers bill. But cried with laughter with many others.
The point is that, for Openai, the most important product is not Sora 2, which is a technological feat but is pending to climb in a profitable way to the rest of the world. But Sora, the social network. Because if Sora manages to convince millions of users and shows them advertising, it will be a crucial source for the company that makes Chatgpt, one that will also be a direct firecracker to the finish line.
So, yes. Openai wants these Sam Altman videos in uncomfortable situations to be content that we consummate for many hours. And that the people who had not created content so far, had only consumed it due to lack of resources to record OX reason, become active creators thanks to the benefits of Sora 2, the model.
Can I imagine in a Scroll infinite seeing content generated by AI when they pass a few days after the initial boom? If I keep having fun, then. It’s funny, it’s creative and has a viral potential that reminds me of Vine’s best days. I am also exploiting the part of creating private memes with my friends using our own cameos.
Do I worry that I am having fun with a content that, although it is original, would not have been possible without television programs, YouTube videos, the Tiktoks and the Reels that OpenAi has surely used to train Sora 2?
It is possible that this same article is used to train future OpenAi models without intermediate agreement, so there is a much deeper and more urgent background debate than Sam Altman’s memes. But there is no doubt that that little man with blue eyes who has collapsed my Feed It knows how to pack and launch products. Especially if you manage to annoy homologists like Mark Zuckerberg along the way.
Images | Sora
In Xataka | YouTube is becoming something much more important for Google than its video platform: in your search engine
GIPHY App Key not set. Please check settings