Preface
As of the date of publication, sora has not yet opened to public testing, nor has it provided an application path for internal testing. Especially in China, users have no channels to participate in trials. Globally, only about 1,000 early OpenAI users have been granted permission to use Sora. openai owns the ownership of these user-generated video content and any form of public distribution is strictly prohibited.
We remind users to be more vigilant and not to be deceived by paid services that may appear in the domestic market claiming to provide sora trials.
Text
In the early morning of the seventh day of the Lunar New Year, I was woken up by two kicks from a child outside the window in my hometown. When I got up and looked for water to drink, I glanced at my phone. The message that popped up surprised me so much that I couldn't sleep all night. Yes, that's right, it's AI, it's openai, it's another epoch-making application, and its name is sora. The
picture comes from the Internet
sora, openai's latest text-to-video model, can generate up to one minute of video content based only on prompt words, text instructions or static images input by the user, with excellent visual quality. These videos not only show detailed scene reproductions, but also vivid character expressions and complex camera dynamics.
picture comes from openai official sora-demo
picture comes from openai official sora-demo
Although sora is not the first text-to-video (t2v) application, it has significantly surpassed it in terms of actual rendering effects and richness of application scenarios. Other tools in the same field have caused a "severe blow" to the AI video industry.
pictures are from the official openai sora-demo
According to the 48 video demonstrations released by the openai official website, sora is not only extremely accurate in the presentation of details, but can also create emotional characters, specific styles of activities, and accurately reproduce the complexity of the theme background. Scenes. What's even more awesome is that this model can not only understand user requests, but also gain insights into how these contents exist in the real world.
picture comes from openai official sora-demo
picture comes from openai official sora-demo
This progress marks that if gpt can understand the human world through language, sora has expanded this understanding ability to the video field.
Just when I was surprised that OpenAI could really hide a nuclear bomb, I didn’t expect Sora to be named by CCTV. Since GPT became popular, CCTV’s reports on overseas AI technology have been relatively conservative. However, the emergence of sora made CCTV uncharacteristically not only report it, but also hailed it as "the first large-scale video generation model", which shows that sora's strength has shocked the world. The
picture comes from the Internet
So what are the advantages of Sora compared to competing products in the AI video field? What are the advantages compared to runway ml, picture.ai, and nvidia's video-to-video synthesis? Why do so many AI video practitioners say they will be laid off immediately overnight? The
picture comes from the official openai sora-demo
The key point is that sora has moved towards the ultimate goal of AI technology - creating a "world simulator" (world simulators). Unlike traditional dynamic frame-filling technology, sora is the first AI that can deeply understand the real world and convert it into video content. This high level of understanding and expression of the real world is Sora's unique feature. The
picture comes from the sora official document
. The most important point in my opinion is that the 60-second generation time is really exaggerated. This is the longest time provided by the popular runway gen-2 ai video generation technology not long ago. Compared to the 16-second duration, this is a significant leap forward. This increased length is not just a win in terms of computing power, but also demonstrates Sora's autonomy and creativity in understanding the semantic content provided by the user, allowing it to create longer videos that deeply interpret and represent a given topic. theme.
picture comes from openai official sora-demo
picture comes from openai official sora-demo
This is a subversive existence, different from the diffusion technology that traditional AI video generation tools rely on - this kind of technology basically combines multiple real Pictures are used to create videos, and the results generated often lack deep meaning and are more like slideshows.
However, sora uses the integration of llm (large language models) and diffusion technology, using spacetime patches processed by the transformer architecture as its core technology. This means that sora can not only understand natural language and interpret text information, but also combine it with its insights into the natural world to break out of the limitations of 2D images and simulate scenes and experiences close to the real world. The
picture comes from the sora official document
. This is a bit like a chef cooking. Before cooking, he cuts various ingredients (video source files) of different sizes into uniform small pieces (represented by patches). Then, he will reorganize these small pieces according to the relationship between time and space (space-time elements) like a disordered puzzle to ensure that the cooked dishes (video content) are both in line with the taste of the diners (user input) and good of plating (visual presentation).
ai generates
. The ability displayed by sora is transcendent. What impressed me deeply is that in its demonstration video, there is a scene transition effect imitating "Need for Speed" that is particularly shocking. In this video, Sora not only accurately captured and transformed the seasons in the original video scene, but also skillfully handled ray tracing to achieve a near-perfect effect.
picture comes from openai official sora-demo
picture comes from openai official sora-demo
picture comes from openai official sora-demo
We know that nvidia has invested in the research and development of several generations of graphics cards in order to achieve realistic light tracing effects in games. However, sora can easily deduce shocking real light effects through simple instructions without traditional scene construction and cumbersome interactive debugging. This technological advancement is undoubtedly a major transcendence of traditional graphics rendering methods.
picture comes from openai official sora-demo
Although sora is not currently in public beta, but just through the released trailer, we can analyze its impact on the vision industry on the day it is launched. The first one to bear the brunt is the video production industry. Video The production process, which originally required a lot of time in script creation, scene construction, shooting and post-processing, can now be greatly simplified and even automated through Sora's AI technology. This could lead to video production becoming more cost-effective and efficient, but would also pose challenges to the careers of professionals working in video production, such as directors, cinematographers, editors and visual effects.
ai generates
Imagine that you are an advertiser of outdoor wedding photography, and your goal is to create a short film that will inspire potential customers to purchase by showing romantic scenes of couples taking wedding photos at outdoor attractions around the world. According to the traditional production process, actors and the filming team need to be arranged to travel around the world and shoot on-site in famous scenic spots one by one. However, with sora, this becomes easy. Just enter a command, sora can generate a beautiful advertising short film for you that is comparable to shooting on the spot, saving time and reducing costs, and you can continue to make adjustments if you are not satisfied.
ai generates
In the past, when faced with reports that lacked real-shot material, media organizations often relied on 3D animation to assist explanations. This process was not only costly, but also time-consuming. Now, just provide sora with a summary of the news and what happened, and it can quickly generate a video of 60 seconds or more. This not only greatly improves the efficiency of news production, but also enhances the attractiveness and vividness of reports.
ai generates
If sora can be implemented in the future, it can overcome the problem of computing power requirements, face more ordinary users, and further adjust the video content according to user needs, such as changing faces, changing clothes, changing scenes, etc., and can properly handle and Copyright-related commercial application issues will mark a new era in digital video production. Currently, Sora still faces many problems, the biggest of which is how to solve the huge computing power demand after public use.
ai generates
This may also lead to the vigorous development of cloud computing and distributed computing.
In any case, OpenAI's Sora allows us to see the dawn before the arrival of AGI. In the future, the maturity of AI technology combined with the continuous evolution of VR AR equipment will gradually move us towards an unprecedented era. In this era, everyone can use AI technology to enter virtual spaces designed entirely by their own imagination. These spaces are not only full of realism, but can also respond to their own feelings and emotional changes in real time, creating their own AI world.
ai generates
