Before the end of the first month, three major events occurred in the science and technology world.
First, NVIDIA launched Chat with RTX, which turns everyone's computer into a large model of a localized system;
The second is that Google AI launched GEMi1Version 5, compared to GPT-4, can handle the input window of more than 1 million tokens; These two things seem to be still a certain distance from the daily life of ordinary people.
But the third thing has blown up everyone's circle of friends - the release of Sora. It is a new generative AI model launched by OpenAI.
In the early morning of February 16th, OpenAI released a blockbuster update and launched the first Wensheng ** model Sora. With text commands, Sora can directly generate up to 60 seconds of **, including detailed backgrounds, subjects, flexible multi-angle shots, and multiple emotive characters.
In just 2 days after the release of SORA, it quickly became the focus of heated discussions on the global Internet, and there were endless reports about it that would completely change the film and television industry and the short industry.
SORA is not yet available to the public, and according to MitTechnology Review, OpenAI currently has no plans to release SORA to the public, and only relevant programmers, security testers, and a small number of creators and artists selected by OpenAI will be able to use SORA.
In addition to dealing with the risks of regulation and potential harm, OpenAI's immediate priority is to put SORA in the hands of visual artists, designers, and filmmakers for testing. It can be seen that this set of Wensheng** model will give priority to film and television-related industries, and OpenAI also hopes to obtain relevant feedback through the evaluation of professionals to promote the progress of the model.
Previously, generative AI has gradually reshaped the way industries such as advertising, finance, and education operate, improving productivity and decision-making by leveraging automation, personalization, and optimization technologies. The birth of SORA means that the era of AIGC (AI-generated content) may have arrived, and tools such as SORA may also profoundly disrupt the future of content creation and business.
We noticed that OpenAI has updated 48 demos generated by Sora, and after our repeated analysis and analysis, we have come to the following opinions:
Sora stands out because it overcomes "physical difficulties".
Compared to the 10-second creation limit of similar products, SORA generation** is up to 60 seconds, and the detailed picture of the generated content has reached a level where it is indistinguishable from the real world. In addition, the consistency between the subject and the background environment of the SORA-generated content is even more impressive.
But the most important thing is Sora's mastery of the world model. Through learning, SORA can understand the real-world operation knowledge and physical laws. However, the previous diffusion model can only realize the ordinary conversion of text and 3D model, and cannot be directly embedded in a virtual physical world.
To put it simply, in the past, the use of Wensheng ** gave people the feeling of being more like "moving**", in which ** there was a lack of dynamic interaction between the background and the subject, and it was impossible to cross the threshold of "real".
For example, the water surface fluid dynamics and the physical difficulties of the scale of the movement have been solved.
Jim Fan, a senior research scientist at NVIDIA, went so far as to say that "SORA is a data-driven physics engine" and "is a learnable simulator, or model of the world ".
2.Sora can level up frighteningly fast
The capabilities of AIGC tools such as SORA are based on big data training, and like the previous ChatGPT, they have a network effect that makes their iterative evolution extremely fast and cannot be accurately measured by Moore's Law.
However, unlike ChatGPT, the AI content generated by SORA is more intuitive for ordinary people, and it is easier to get feedback from a large number of users and practitioners. Since short** is the mainstream entertainment and information ** in the current world, its wide range of information ** and communication channels provide strong support for SORA's self-learning and improvement.
Although there was also a misunderstanding of physics in the early days, such as juice spilling from the bottom of the cup, it can also show that SORA is constantly exploring the physical world independently, and this upgrade method through error correction and learning is more in line with people's understanding of "intelligence" in the future.
Therefore, SORA's understanding of the content created will continue to upgrade, and the comprehensive level has greatly surpassed the previous popular Runway and PIKA, with the continuous accumulation of big data, SORA's development prospects are currently not endless.
Zhou Hongyi, the founder of 360, said that once Open AI's artificial intelligence watches all the movies and **, this is really not far from AGI, it is not a problem of 10 years or 20 years, and it may be realized soon in one or two years.
The quality of the creation of the average person using SORA is not inferior to that of most professionals
Once upon a time, shooting a single shot required the purchase of an expensive camera and learning to adjust various parameters, while rinsing involved complex chemistry and multiple steps. However, with the advent of digital cameras and smartphones, every ordinary person can shoot anytime and anywhere, which makes the professionalism of traditional photographers gradually only show through abstract factors such as composition, lighting, and personal style.
Now, SORA is about to make that change. With the improvement of its large model and the enhancement of its self-learning ability, Sora can allow an ordinary person with no experience in film and television production and art design to directly generate a ** that meets his description, and the effect brought by this ** is comparable to the special effects produced by the high-cost production of sci-fi blockbusters.
This means that future online writers may also complete a ** adapted fantasy short film on their own while writing articles. And for professionals in special effects production and virtual set construction, if the high-cost production of the picture ends up being similar to the work of grassroots authors, it will undoubtedly put a lot of pressure on them.
For short** creators, the value of authentic and emotional content will rise infinitely.
Short** creators should think deeply about the fact that in the context of SORA and other tools to promote the development of AIGC, short**, as a mainstream information acquisition channel, will be filled with a large amount of AI-generated content, and the classification of live, landscape, narrative and other ** is the "hardest hit area" of generated content.
And with the continuous advancement of AIGC technology, it has become difficult for users to distinguish whether the content is created by AI, which makes it less important for users and creators to distinguish the authenticity of content.
But no matter how much SORA "understands" the world, and no matter how realistic the content it generates, they are always the product of digital simulation, and cannot replace the reality shown by real shooting. Therefore, in order to avoid being overwhelmed by the torrent of generated content in the future, short creators will achieve better results if they focus on real shooting and touch the audience through deep emotional display.
Since the release of SORA, the film and television industry has first received a lot of attention. In particular, the special effects generated by SORA are the most amazing, and they are not inferior to Hollywood blockbusters. Many people believe that SORA can reduce the production cost of visual effects in the film and television industry, thereby changing the production model and industry chain of the film and television industry.
Our team also quickly shared a few paragraphs** generated by SORA with practitioners in the film and photography industry. Several visual effects experts said that for the ** of the realistic class, the generation effect of SORA is not real. The average person may only find it a little weird, because most of the distortion problems are in the light and shadow.
For the CG (computer animation) small scenes** generated by SORA, they said that they are in place overall, and even the works of many related CG production companies cannot be compared with SORA at all.
While Sora excels at generating CG and producing stunning 60-second** content at a fraction of the cost, a movie isn't simply a patchwork of 60-second segments.
The film and television industry often needs to produce large, coherent scenes, which is not yet available to SORA, and the content generated by SORA is difficult to stand up to professional scrutiny in terms of detail.
The 60s short ** has completely different narrative requirements than the long ** movie, similarly, AI can write a good joke, but it is difficult to write tens or millions of words excellent**.
We believe that despite Sora's strong ability to generate and learn, it is still difficult to express a lot of content and detail in scenarios. At present, SORA cannot completely replace manual work to take over the work of film and television creation. However, there's no denying that Sora is an excellent tool for concept ideas. In particular, its advantages in small scenes** indicate that it will have a profound impact on the advertising industry and the creative industry.
Another area that has attracted a lot of attention is the platform. The data shows that the current scale of domestic short-term users is more than 1 billion, of which Douyin's annual revenue in 2022 will reach more than 70 billion US dollars, and Kuaishou will also achieve an annual income of 90 billion yuan.
However, in the face of such a large market, all the ** generation tools failed to meet the standards of commercial or industrial production before the release of SORA.
Compared to Pika and Runway, Sora not only provides beautiful picture quality, but also has richer and more diverse content, and at the same time, the duration has increased by more than ten times. If it is used for short** creation, this will greatly enhance the freshness of the user. However, when many creators choose to use SORA to output content and thus passively "homogenize", how to ensure the quality of their works is outstanding, which is the real problem of using SORA.
The rise of AIGC has lowered the threshold for revitalization, hot topics and jokes, resulting in relying solely on generation** is not enough to maintain the competitiveness of creators. At present, it seems that the narrative self may be more able to give full play to the advantages of SORA, because SORA has not yet been able to generate ** with a unique tone and core creativity.
Therefore, for content creators on the platform, SORA is not a substitute for their own creativity and inspiration, but can only be used as an auxiliary tool to improve the efficiency and quality of creation.
As far as the ** generated by SORA so far is concerned, the excellent works are mainly concentrated in the fields of animals, long-range architecture, and fantasy scenes. In the past, these often required creators to pay high royalties to use them. Therefore, the emergence of SORA is likely to disrupt the material copyright industry in the first place.
SORA's powerful generative power is accompanied by a potentially huge destructive force to the social order, so it will inevitably attract some people with malicious intentions to use it to commit fraud, blackmail, slander and other illegal acts.
It is foreseeable that the official launch of SORA will not only face its own iterative optimization, but also face strict supervision in many countries and regions around the world, and we believe that OpenAI will not release SORA to the public in the short term. But no matter when Sora is released to users, it will be further proof that AI has become an integral part of human society.
The emergence of SORA is undoubtedly a great success in AI development. It is based on the existing knowledge base of human beings and the world model, and superimposes relevant self-learning technologies, which is undoubtedly one of the right paths for AI development. AI companies will easily use this model to build super tools across industries.
In the past few years, concepts such as the metaverse, VR, and artificial intelligence have sprung up, but there has never been a concrete product. Today, Meta headsets have tens of millions of sales; Nvidia has reached a market capitalization of 1$7 trillion; Apple has also released its own headset, the Apple Vision Pro; OpenAI's artificial intelligence products are also constantly being updated. It can be seen that an era around virtual and artificial intelligence is coming to us with an irreversible posture.
end-