Written by: Zheng Minfang

Source: Wall Street Journal

When OpenAI across the ocean seems to have pressed the "pause button" on the AI generation model Sora, Chinese tech giants have launched a counterattack in this field.

Recently, ByteDance launched its latest AI video generation model Seedance 2.0, which quickly exploded on the internet thanks to its advantages in multimodal input, automated camera movement, and consistency.

Feng Ji, founder of Game Science, made a weighty judgment after an in-depth experience: "The content field will surely face unprecedented inflation."

Feng Ji's prophecy is not alarmist.

This shockwave is rapidly transmitting to industries such as e-commerce, gaming, video platforms, and film production: in e-commerce, the technical threshold for low-end outsourcing and shooting bases has been completely flattened; in the gaming industry, the production cycle for concept verification and promotional materials has been drastically compressed, leading to more brutal competition; video platforms must further optimize distribution logic to cope with the explosive supply; and the traditional linear process of film production "shooting + editing" is facing a dimensionality reduction blow from the "prompt + generation" industrial pipeline.

A major industrial reshuffle regarding benefits and replacements has already begun.

The Explosion of Video Production Capacity

In the past year, the biggest pain point for AI video has been deliverability.

Whether it's Sora, Runway, or domestic Keling, or even ByteDance's self-developed Jimeng, all exhibit this problem. Creators often get stuck in a "drawing card" game, having to generate dozens of times to obtain a few seconds of video that is not flawed and maintains consistency.

The core breakthrough of Seedance 2.0 lies in attempting to transform "showing off techniques" into "deliverable narratives."

The breakthroughs in key capabilities are mainly reflected in three aspects:

First is multimodal input. According to tests by All Weather Technology, new members logging into Jimeng only need to pay 1 yuan for automatic renewal to directly use Seedance 2.0, supporting text, images, videos, and audio as reference material inputs, meaning almost any format you can think of can be input to generate videos.

Second is understanding narrative and learning camera movement. Seedance 2.0 displays "director-level" thinking, not only able to understand complex narrative logic but also automatically scheduling cinematographic language, completing operations such as zooming and panning. Videos are no longer simple shifts of static images but possess cinematic narrative logic.

Third is the consistency of images. According to tests by All Weather Technology on various AI video generation applications available on the market, issues such as facial expressions collapsing during movement and backgrounds appearing with clear blurriness have been rampant.

However, from the demo video, Seedance 2.0 maintains the consistency of facial and visual information during the subject's movement, making coherent narrative expression possible.

This means that AI video generation is transforming from a toy into a tool. The ability to standardize video generation industrial pipelines makes "everyone a director" no longer an empty slogan, and this will significantly compress the costs of video production.

Feng Ji described this transformation as "inflation."

"The production costs of general videos will no longer adhere to the traditional logic of the film industry, gradually approaching the marginal cost of computing power. The content field will surely face unprecedented inflation, and traditional organizational structures and production processes will be completely restructured. I believe that anyone who has used it will quickly understand that this prediction is not alarmist," Feng Ji stated.

The First Wave of Impact

As the marginal cost of video production approaches zero, the business models built on the old cost structures will be the first to be impacted.

The four major industries of e-commerce, gaming, video platforms, and film production may be the first affected.

The most immediate impact is first felt in the e-commerce sector.

Product presentations, scene interpretations, and functional explanatory videos do not essentially rely on complex artistic narratives but on clear information transmission.

With the popularity of Seedance 2.0, the threshold for businesses to obtain video expression capabilities has been completely flattened. Low-end video outsourcing companies and Taobao shooting bases that previously survived on "information gaps" and "technical thresholds" will face a harsh winter, and video production may shift from professional outsourcing services to businesses' daily operations.

Compared to e-commerce, the impact of AI video generation models on gaming might still be relatively limited, but the revolution has quietly begun.

The costs for worldview interpretations, concept verifications, and promotional material videos are falling exponentially. More projects will be validated at earlier stages and eliminated earlier as well.

An insider from a Beijing game company told All Weather Technology that the company has already initiated small-scale testing of Seedance 2.0 internally.

AI video generation models are also changing the distribution logic of video platforms.

For platforms like Douyin and Kuaishou, videos generated by models like Seedance 2.0 bring a surge in content supply, forcing the platforms' core competitiveness to completely shift to "selection and distribution" mechanisms, such as whose algorithm can more accurately sift gold from the massive sea of AI-generated content, and whose commercial conversion efficiency is higher, that will be the winner.

In the film industry, Seedance 2.0's multi-camera narrative capabilities may reshape production processes.

In the past, the birth of a film or television work often followed a strict linear industrial process: first, there was a large volume of footage shot, then editors would make selections and assemble them in post-production, constructing narrative logic.

However, within the logic of Seedance 2.0, this boundary is becoming blurred.

In the filming stage, there is the possibility of low-cost scene generation by AI models; the model itself has an understanding of camera movement and narrative rhythm, and by the time it's generating a video, it has actually already synchronously completed the "editing" work.

AI is no longer just spitting out scattered material shots but delivers a "complete film" with coherent temporal and spatial relationships directly.

This means that the traditionally time-consuming post-editing phase in film production faces the risk of being "dimensionally reduced" by algorithms.

Future creation flows may no longer be "shooting + editing," but rather "prompt + generation," shifting editors' roles from "operators" to "instruction engineers" or "aesthetic gatekeepers."

Although the videos generated by the current Seedance 2.0 are not perfectly flawless, with logical details and images still needing improvement, the speed of technological iteration far exceeds market expectations, and these issues will not hinder progress in the near future.

The "Moat" of IP

Seedance 2.0's astonishing "replication" capability has allowed ordinary people to enjoy the joy of creation while also placing unprecedented pressure on rights holders.

Recently, a large number of "recreations" and even "parody" clips of Stephen Chow's classic films have gone viral on short video platforms.

With the computational power of AI video generation models, Stephen Chow's facial expressions, signature laughter, and even classic dialogue styles have been low-cost replicated by numerous users, generating absurd plots that never occurred.

This quickly drew attention from Stephen Chow's team.

Stephen Chow's agent, Chen Zhenyu, publicly questioned in a post: "I want to ask, are these infringing (especially with large-scale dissemination these past couple of days)? I believe the creators must have profited, and is a certain platform just allowing this user-generated content to be published without any oversight?"

This questioning seemingly unveiled the copyright anxiety of the AI era, but from a deeper commercial logic perspective, it actually corroborates the extreme scarcity of top IPs in the AI era.

In the future, amidst the flood of AI-generated content, technology itself will no longer be a barrier, as everyone will have access to the same Seedance 2.0 tools.

The true barrier still lies in the hands of IP owners.

Precisely because the market is flooded with many "high-fake" Stephen Chow replicas, the irreplaceable nature of "real Stephen Chow" IP is further highlighted.

When content supply is not only excessive but also "inflating," user time and attention will become exceedingly precious. Those able to instantly capture users' attention will remain classic IPs that have been validated by time and possess strong emotional penetration.

In other words, although AI has lowered production thresholds, it has infinitely heightened the value of "recognizability."

For IP owners, the future remains bright. Long-accumulated IP assets will no longer merely be objects of infringement but can realize exponential commercial value through legitimate licensing under the leverage of AI by countless creators.

From the launch of Sora 1.0 by OpenAI in February 2024, becoming the world's first AI video generation model supporting up to 60 seconds of generation, to the present where ByteDance's Seedance 2.0 achieves multimodal input to generate 60 seconds of original audio narrative blockbusters, it has only taken 2 years.

In this rapidly advancing technological era, various industries stand at a crossroads: the costs at the execution level are being infinitely compressed, and those repetitive roles reliant on manpower and time will be ruthlessly replaced; meanwhile, the value of IP and creativity is being infinitely amplified.

When tools become readily available, the height of content will no longer be determined by whether one knows how to use the software, but rather by whether the conception about the world in one's mind is unique enough.