With the arrival of Sora, whose job will it take away?

OpenAI has a new life. In early 2024, OpenAI released a new video generation model Sora, which attracted widespread attention from various industries. This article shares the advantages of Sora compared with other video generation models, and infers the impact that Sora will have on the film and television industry. Come and read it!

Following ChatGPT, OpenAI sparked another discussion in early 2024.

On the morning of February 16, OpenAI released the Vincent video AI model Sora, which set off a global Internet sensation. Unlike previous Vincent video models Runway and Pika, Sora can continuously simulate people, animals, and objects, and generate multiple shots of the same character, maintaining its appearance and background throughout the video.

Sora can also generate images with more detail than ever before, including spots on the face and fine lines on the lips, at resolutions up to 2048×2048.

After Sora was released, many tech leaders came out to discuss the topic. Musk said "gg humans" (gg/good games originally refers to the greetings between players at the end of the game, and later extended to "game over"); Zhou Hongyi predicted: "This may bring huge disruption to the advertising industry, movie trailers, and short video industries"; former Alibaba vice president Jia Yangqing commented: "It's really amazing."

What are the highlights of the Sora model, which has dominated the technology sector recently? Specifically, what impact will it have on the film and television production industry?

01 High quality, long duration, multiple scenes

Simply put, Sora can create 60-second videos based on text prompts, extend existing videos, and generate videos from images, including complex scenes and camera movements.

OpenAI's official website shows several videos produced by Sora. A woman in a black leather jacket and a red skirt is walking on the streets of Tokyo at night after the rain. The dark pores on the woman's skin are clearly visible, and the water on the road reflects the reflection of the street lights. The video is very realistic. If it weren't for the occasional exposure of the left and right legs, it would be difficult to tell at a glance that it is a video produced by AI.

In terms of camera movement, composition, etc., Sora videos have shown significant improvements, bringing AI videos to the "next level" in one fell swoop.

Image source: OpenAI

In addition, Sora can not only generate a full 60-second video, but also extend the generated video. In other words, if you give Sora a video, it can automatically generate the previous or next video.

What's even more outrageous is that Sora can generate videos with different camera positions and different angles, and edit them. And under different camera positions, whether it is wide angle, medium shot, close shot, close-up, indoor or outdoor, the relationship between the characters and the background in the video is consistent and unaffected.

Image source: OpenAI

In other words, with just a paragraph of text, the Sora model can generate a 60-second 1080p video containing different shots. It makes people wonder: "How big is the gap between reality and fantasy?"

It is worth noting that Wensheng Video has already existed. According to statistics from the well-known investment institution a16z, by the end of 2023, there were 21 public AI video models on the market, such as Google's Lumiere, Stability AI's SVD, and Runway, the developer of the video generation model Gen-2. Among them, Runway completed its Series C financing at the end of June 2023, with a valuation of more than US$1.5 billion.

After Sora was released, Dongwu Securities compared the main video generation models. He compared and analyzed the characteristics of the six models, including Sora, WALT, Gen-2, Emu Video, Pika 1.0, and Stable Video, as well as the performance of the generated videos. The conclusion is that Sora has significant advantages in terms of generation time and consistency, and has a breakthrough semantic understanding ability.

Image source: Soochow Securities

At the beginning of last year, ChatGPT was born, and a year later, Sora achieved the rapid creation of videos. Such a fast development speed is shocking. After all, a year ago, AI generated videos were still like this.

Image source network

A user on Bilibili said: "When I was a kid, I wondered if there would be a technology for making movies in the future, where people would wear brain machines and use their imagination to generate all kinds of magnificent movie scenes. Who knew that this reality is not far away."

Although Sora is still in the testing phase and is only open to invited creators and security experts (allegedly some visual artists, designers and filmmakers), the capital side has already heard the news and moved. Data from CB Insights shows that OpenAI is currently one of the most valuable technology startups in the world, second only to ByteDance and SpaceX.

After the sale of existing shares under the acquisition offer led by Thrive, OpenAI's current valuation has reached more than $80 billion, nearly three times what it was nine months ago.

However, some are happy while others are sad. For some film and television industry practitioners and AGI video start-ups, the advent of Sora can hardly be said to be a happy event.

02 The Storm

The most direct impact of Sora’s release is on the AGI video startup.

Runway, which participated in the production of the 2023 hit film "The Blink of an Eye", posted two words on the X platform after Sora was released, "Game On." (The competition has begun).

Image source: X platform

For ByteDance, the emergence of Sora is undoubtedly a major threat to Jianying. This year, just one week before Sora came out, Zhang Nan, the former CEO of Douyin Group, resigned and turned to Jianying, reflecting Douyin's emphasis on AIGC tools. With the continuous development of Sora, how Jianying can learn from Sora and innovate has become a top priority.

In addition, Sora-type AI models have the most direct impact on Hollywood and the fields of film, television, advertising, etc.

A survey of 300 Hollywood industry leaders released last month by U.S. industry research firm CVL Economics showed that 75% of respondents admitted that generative AI (tools, software, models) had prompted them to cut and merge jobs in their business departments, and a mood of concern was spreading throughout Hollywood.

Those bigwigs who control the industry order in Hollywood predict that in the next three years, a total of more than 200,000 jobs in Hollywood will be impacted by AI, especially post-production jobs such as visual effects, sound effects engineers, and painters.

Image source: OpenAI

However, looking back at the history of content creation, the development of tools is unstoppable and progress is the norm. Instead of resisting, creators should think about which links and content are becoming more valuable.

From the perspective of the AI ​​video production process, the current Sora needs to input a text first and then generate a video. The originality of the video still depends on the creator's aesthetic taste. Sora's tool attributes are more prominent. Compared with original content, Sora's advantage lies in those special effects clips that require a lot of manpower and material resources.

Therefore, some netizens predict that although editors, special effects artists and other post-production positions in the video production process will face more severe situations in the future, the content that was limited by shooting costs and shooting technology in the past will receive more attention.

Ideally, Sora will be able to replace more mechanical and repetitive work in the future, allowing creators to focus on innovative and in-depth interpretations and provide cultural consumers with better content.

In addition, since AI's understanding of content is more inclined to input "keywords" rather than scripts, in the future, how to create scripts suitable for AI to understand and generate videos is also a problem worthy of attention.

03 Sora’s value is more than just video

At present, Sora's most direct impact is the video production industry, but his ambition, or the ambition of many large models, goes far beyond that.

OpenAI's official website positions Sora as a world simulator. OpenAI believes that it can effectively simulate the physical and digital worlds, including objects, animals, humans and other factors. According to OpenAI's report, Sora has made great progress in understanding the laws of the geophysical world.

Of course, Sora as a simulator currently has some flaws, and its world model is still not perfect. In the 48 Sora-generated videos released by OpenAI, there are many mistakes.

For example, the glass has not broken, but the liquid has already flowed out; people dug out a deformed plastic chair in the desert; a man ran on a treadmill backwards, and other illogical video contents. In short, some causal laws that are conventional for humans cannot be inferred by the Sora model in the short term.

Image source: OpenAI

Based on the information available, Sora is still in the 1.0 stage and often has difficulty processing detailed backgrounds, but no one will deny Sora's milestone status on the road to achieving AGI.

In the AI ​​boom, the emergence of Sora has shown us the possibility of realizing AGI, and has also forced the industry to continue to innovate and develop. After all, after the bubble, there can only be one winner.

Author: Guang Ye

Source: WeChat public account: TopKlout (ID: TopKlout)