China's top 3 players in the video model sector have embarked on divergent competitive paths.

Seedance focuses on industrialized production, Kling prioritizes controllability, while Vidu delivers unrestricted generation and free expression.

Around the National Day period, a series of consecutive moves by Seedance, Kling and Vidu have made the competition of Chinese video models not only stay far ahead globally as always, but also explore multiple development paths under a tripartite pattern:

Some players focus on the industrialization efficiency of content production, some bet on the control capability required for professional production, while others try to integrate the expressive power of characters, camera movements and visual effects, together with lower trial-and-error costs into one single product.

This path differentiation is soon reflected in specific product launches.

On September 28, Kling AI officially announced its new-generation video generation model Kling 4.0; on October 8, Shengshu Technology released the first preview version of its new-generation flagship video generation model, Vidu Q4 Preview, with pricing starting from 0.09 yuan per second, further pushing the generation effect and cost performance to the extreme.

After the model was released, many professional video directors conducted in-depth experience tests. For example, content creator Senhai Yingguang used Vidu Q4 Preview to produce the short film "Green Mountain", which tells a story of an old general guarding the city who exchanges his own life with the attacking general for the safety of the people in the city.

In the past, AI video was most prone to two types of mismatches: the characters look exquisite but have no emotional progression; the scenes look lively but the camera movements have no narrative purpose.

The short film "Green Mountain" provides a very good observation sample. The prominent features such as the white hair, wrinkles and beard of the old general in the film provide visual anchors for continuous character consistency;

The large scenes in the film, such as marching in valleys, Great Wall passes, and military camps at night, have unified art styles, and the low-saturation tone and partial candlelight processing are very cinematic.

This short film produced by Vidu Q4 Preview not only proves that AI is already capable of making films, but also perfectly illustrates how we should measure the capability of video models: whether characters, scenes, camera movements and emotions can move forward with the story together.

This is exactly where "vivid expression" is far more difficult than "clarity". A pause in a gaze, a pullback shot after a turn, a brief silence before an explosion, all determine whether the audience can understand why this character behaves in such a way.

In addition to model performance, what is more noteworthy are the signals of price and efficiency: the invocation cost of flagship video generation models is entering a range that ordinary creators and small teams can use at a higher frequency.

This indicates that in the process of Chinese video models evolving from technical competition to production competition, model and engineering efficiency are being converted into lower creation costs.

Is the "DeepSeek moment" for video generation models really coming?

01. The track competition is heating up, Shengshu Technology enters the top camp

The AI video generation model track in China is undergoing explosive development.

Data from LeadLeo Research Institute shows that between 2020 and 2025, the market size registered a compound annual growth rate of 965%. In 2026, the market size reaches about 3.66 billion yuan. It is estimated that by 2030, this figure will jump to nearly 149.7 billion yuan, with an average annual compound growth rate of 152.9%.

Against this booming background, the industry pattern is accelerating its transition from "hundred flowers bloom" to "top players gathering". Players including ByteDance Seedance, Kuaishou Kling and Shengshu Technology have formed the line-up for the "final round" of the track.

The term "final round" itself carries the connotation of competing on the same field, which reminds me of the Chinese national swimming team.

After Sora was shut down, domestic models have successively broken the "long-standing records" that were long dominated by international giants in the window period. With different technical routes, all of them have achieved world-class results in the same pool, jointly lifting the overall industry level.

The core advantage of Seedance lies in its backing of ByteDance's complete content ecosystem.

Seedance 1.0, launched in June 2025, once ranked first in both the text-to-video and image-to-video rankings of Artificial Analysis. The release of Seedance 2.0 also sparked widespread cross-circle discussions, which fully proves the solid foundation of the model. It currently leads the market with absolute advantages, and has verified to the whole world that video is the second large-scale commercial implementation track after coding.

The advantage of Kling lies in its in-depth cultivation of vertical scenarios and rapid commercialization. Kuaishou deeply binds Kling to its own advantageous businesses such as short dramas, e-commerce and live streaming, with particular focus on the film sector. These scenarios are not only product polishing grounds, but also sources of revenue.

In the "final round" line-up, Shengshu Technology is one of the few startup players. It is also an AI Native player with formidable technical strength that cannot be underestimated, especially considering its intelligent reserve potential in the development of AGI. After all, in the development of the AI video sector, this team has created several milestone achievements.

From its initial focus on long-duration, high-consistency generation, to the early launch of reference-based video generation in the industry - upload several reference images, and the model can maintain the consistency of characters and objects to generate new frames - and then to aesthetic-demanding themes such as animation, Vidu has been proving one thing: startups can also stand in the first echelon in terms of generation quality.

This time, the new Q4 preview version has surpassed Seedance 2.0 in the AA ranking, ranking third in the world. The previous Vidu Q3 once ranked first in China and second in the world with a score of 1241, surpassing Runway Gen-4.5, Google Veo 3.1 and OpenAI Sora 2 in one go. The official Q4 version is definitely well worth looking forward to.

With industrial-level expressive power, trial shooting entering the sub-second era and ultra-high definition entering the angle era, the industry inclusive driving force of the flagship video model allows Vidu to broaden its own competitive space beyond Seedance's ecological barrier and Kling's scenario in-depth cultivation.

The release of Vidu Q4 Preview is a further reinforcement of this differentiated positioning.

02. Vidu Q4 Preview has achieved "vivid expressive power"

Vidu Q4 Preview is Shengshu's new-generation all-around flagship generation model with high expressive power, focusing on "vivid expressive power", with upgrades centered on three core capabilities: character performance, camera language and complex visual effects.

In terms of character performance, the generated characters have further improved performance in facial expressions, emotions and body movements. When a character hears a sentence, sees an event or makes a decision, the emotion transmits from the face to the whole body, and the whole process is coherent and natural.

For example, in this case, the male lead shows obvious emotional progression when facing the death of the female lead, with tears bursting out, and the emotion gradually escalates from panic, supplication, grief to complete loss of control, which is highly immersive.

In terms of camera language, Vidu Q4 Preview can generate dynamic camera effects such as follow shot, surround, push in, pull back, lift and lower, and rotation. During high-speed movement, character turning, occlusion and perspective switching, the camera movement and character action cooperate naturally, with almost no jump or spatial disorder.

In terms of complex visual effects, Vidu Q4 Preview can generate effects such as explosion, flame, smoke, fireworks, particles, and energy impact. After the special effect occurs, the surrounding light and shadow, smoke and particles diffuse with the movement, and the character action keeps synchronized with the special effect process, which is very immersive.

For example, in the fast season switching shot, the model can achieve smooth transition of spring petals, summer fireflies, autumn fallen leaves and winter snowfall; various elements can also dynamically change colors following the ambient light, with unified rhythm, and the conversion effect will not appear stiff and disjointed.

In addition, Vidu Q4 Preview supports up to 4K output, and can generate a maximum of 16 seconds in a single run, providing sufficient space for continuous actions, camera progression and emotional changes to unfold.

At the same time, it supports up to 15 reference images and 3 reference audios. Creators can combine visual information such as characters, scenes and products, and then use text descriptions to clarify the relationship between them, adding more details.

For example, users can send character voices, character setting images and storyboard images to Vidu Q4 Preview.

Vidu Q4 Preview can generate a complete manhua drama short film based on the reference images and audio.

03. The unique Vidu style meets the rise of Chinese animation, a historic opportunity

The release of the new vivid Q4 has triggered extensive discussions among users about the unique Vidu style —

The so-called "Vidu style" refers to animation sense, speed sense and explosive power. In the words of experienced users, it is like "igniting imagination and pushing it straight to the sky".

This style is first reflected in actions and camera movements. When the character turns around, the hair strands swing along with the movement; when gold coins fly apart, the particles are directional and hierarchical; the camera chases the character and rushes out, and knows to pause at a single gaze. It not only captures the shape of the action, but also tries to capture the spirit behind the action.

This performance has attracted some creators who used to make abstract art in the Vidu Creator Plan (CPP) to start trying animation narration that they are not familiar with; Japanese AI animation creators are also continuously using Vidu. In horizontal comparisons, people keep mentioning that Vidu is "so exciting".

But "exciting" is only the surface. In this round of video model competition, animation is not just a conveniently added "style pack", but a main battlefield worthy of separate competition.

The reason why it is a main battlefield is that it is connected to greater cultural opportunities. The rise of successful works such as "Nezha" and "Black Myth: Wukong" jointly reminds the industry that Chinese mythology, history and aesthetics can reach global audiences through new visual languages. The next challenge is how to allow more original characters and stories to grow continuously.

The international dissemination of Japanese animation provides a reference: what really brings cultural influence is actually the long-term IP operation supported by characters, serializations, films, games, music and derivative consumption.

China does not need to copy Japanese animation, it only needs to give local legends, sci-fi imagination and contemporary life the opportunity to evolve from a setting and a short film into works that people are willing to follow and share.

To win this main battlefield, it ultimately depends on model capabilities. But at the same time, the animation scenario can in turn fully test all the capabilities that Q4 Preview focuses on: in literary scenes, can a character catch the previous line of dialogue with subtle expressions; in action scenes, can the blocking, chasing and turning movements be coherent; after the shot switch, is the relationship between the character and the space still correct; are the fireworks and energy effects amplifying emotions or covering up actions.

At the same time, the availability of multiple generations, shot connection and post-editing are also put on the table, so that creators can see where the differences lie.

If the model allows small teams to test characters, storyboards and highlight actions at lower costs, what it helps is not only "faster film production", but also enabling more potential original Chinese IPs to cross the first round of trial shooting threshold.

Cultural construction will not be completed by a single model, but giving creators more affordable expression opportunities is a real step forward for this entire industrial chain.

04. From capability breakthrough to creation inclusiveness, Shengshu Technology stands at the critical point

Does the rapid development of technology mean that video generation has ushered in the "DeepSeek moment"? To answer this question, we must first clarify what the "DeepSeek moment" really means.

In January 2025, DeepSeek-R1 was released and became an instant hit. It opened the most cutting-edge model capabilities to the public in an almost free way, promoting real technology equalization. To break it down, this "moment" means the simultaneous occurrence of three things: cutting-edge performance, system efficiency, and mass availability.

Nowadays, video generation models are gradually approaching this moment.

Shengshu Technology has laid out two paths at the same time: MaaS for developers and SaaS for creators, so that different users can make use of the technology.

Vidu Q4 Preview has pushed the threshold down to starting from 0.09 yuan per second: the MaaS image-to-video 720P price is 0.59 yuan per second, the 1080P price is 0.75 yuan per second, and the 4K price is 2.43 yuan per second, all of which are the lowest across the whole network. On the SaaS side, there is a two-month limited-time discount during the Preview stage, and part of the usage cost drops to a few cents per second.

For the industry, the gap between a few cents per second and several yuan per second is the watershed that determines whether large-scale implementation is possible.

The manhua drama industry competes in production capacity. Relying on the mode of high-frequency daily updates and long serializations, the production cost per episode directly determines whether the project can be operated sustainably