Huawei's openJiuwen system provides event-based video memory for agents, reducing response time by up to 97%

ByteDance is restructuring its generative video projects by merging two of its artificial intelligence-based products.

Recently, ByteDance's CapCut division integrated Xiaoyunque into Jimeng, while maintaining the option to access Xiaoyunque as a separate product. Xiaoyunque specializes in automated production of short dramatic storylines, transforming scripts into storyboards and finished episodes. Meanwhile, Jimeng is a broader platform for creating, editing, and generating images and videos.

The functionality of these two products began to overlap as both aim to create end-to-end video content. Now, Xiaoyunque's short drama capabilities have been absorbed by Jimeng, and operations will be managed by Jimeng's Head of Operations and Marketing, although the development team will remain the same.

The main reason for this move is resource concentration. Both products continue to grow, but user growth has not yet generated revenue comparable to investments, and the widespread adoption of video agent products narrows the space for differentiation.

This pressure is also visible in the market: Kuaishou's Kling 4.0, released this month, supports longer generation, multimodal links, and keyframe control. Similarly, MiniMax's H3 is oriented towards commercial use, including advertising, e-commerce, and gaming.

The boundary of competition is shifting from basic text-to-video generation to controlled end-to-end content creation workflows. Xiaoyunque has transformed into a specialized workflow for short dramas, while Jimeng is moving from general image and video generation to professional creation.

ByteDance has also repositioned Jimeng itself. Before the Chinese holidays, the company reduced investment in the consumer application Jimeng and redirected resources to its professional web product, the international version Dreamina, and commercialization.

Thus, the recent consolidation may allow ByteDance to retain Xiaoyunque's specialized short drama capabilities while reducing functional duplication between the two increasingly overlapping products.

A more significant change in the competitive landscape involves Seedance. ByteDance provided Seedance 2.5 via the Volcano Engine API, allowing companies and developers to integrate video generation capabilities into their own products. This model accepts text, image, video, and audio inputs, and also supports video editing and extension.

Volcano Engine documentation shows that the model supports text, graphic, video, and audio input, as well as video editing and extension. Business revenue from Seedance in a Model-as-a-Service (MaaS) mode through Volcano Engine significantly exceeds the revenue from Jimeng and Xiaoyunque.

This means that some applications competing with ByteDance's own content creation tools could also become customers of its underlying model. This raises the requirements for ByteDance's proprietary applications: if Seedance generates revenue through third parties, every internal product must justify its existence through growth, user retention, or monetization.

This restructuring does not look like a retreat from video, but rather a redistribution of ByteDance's bets. Despite consolidating AI tools for content creation, ByteDance continues to expand distribution. Its overseas short drama app, PineDrama, entered the top 10 free apps on iPhone in the US, Brazil, and Thailand, relying on an ad-supported model, traffic from TikTok, and the ByteDance content ecosystem.

These moves may indicate a more multi-layered strategy: developing consumer platforms for distribution, focusing creation tools on professional workflows, and monetizing Seedance by providing its technology to other companies. As AI models improve, creating standalone applications becomes easier, shifting greater power to the companies behind the models. ByteDance's task is to determine where it can still differentiate itself.

The film 'Sanxingdui: Future Past,' which runs for 100 minutes and was created using an end-to-end AI-driven workflow, premiered in Chinese cinemas on October 23rd. According to IT Home, this is the first film of its kind to receive public screening approval from the National Cinema Administration of China.

The film's production was carried out by Bona Film Group utilizing its proprietary Boka Film Cloud Studio platform. Over a period of more than two years, the team generated over 1.2 million base images and conducted more than a hundred full iterations of the film. The creators emphasize that every character is an original digital creation, and there are no digital copies of real actors in the movie.

Bona states that AI is primarily responsible for generation and execution, while ideas, script, shot design, and aesthetic judgments come from the AIGMS team of the company. The production process was organized according to the principles of a traditional film crew, involving about 100 people and 17 specialists, including screenwriters, directors, producers, production designers, cinematographers, lighting, sound, and editing specialists.

The plot unfolds in the future, where humanity has delegated all decisions to algorithms, triggering a global crisis. Clues to resolving this problem lie within seven ancient symbols from the Sanxingdui Bronze Age archaeological site located in Sichuan Province. The creators describe this structure as answers from the past and a crisis in the future.

The film's attendance has been modest. The publication National Business Daily reported low occupancy rates during the initial days of release. Specialists working in the AI content sector told the publication that long films created with AI still face challenges regarding consistency of characters, scenes, and details over 100 minutes, as well as image realism. One expert noted that viewers might perceive the AI-generated characters as artificial, hindering emotional connection. Another argued that the original script determines the final outcome: it is a content market, not a production market.

Bona announced that other AI-involved projects are in development, including the sci-fi animation 'Unplanned: Operation Guanghan,' which has not yet received a screening license. Other Chinese studios, such as Huayi Brothers and Huace, have also announced film and series projects utilizing AI. Currently, the film serves as a test to see if a complete AI-based production pipeline can create a picture that audiences will prefer over traditionally made films.

Dawa Technology, a company based in Chongqing that originated from the field of virtual film production, has introduced Scenes AI—a real-time world model. This system is capable of transforming video shot with a regular camera into an editable and persistent world.

The product received an excellence award at NVIDIA's intermediate startup event in Suzhou on October 15. Thirty members of NVIDIA Inception participated in this stage among more than 580 attendees.

Most video models focus on predicting the next frame. However, Scenes AI supports a continuous world state. Elements such as the scene, people, camera, lighting, objects, and physics are stored as separate variables that can be read and modified independently. When a change occurs, only the corresponding part of the state is updated, not the entire image redrawn.

In demonstrations, Dawa showed how a room transforms into a hand-drawn meadow while the chair, teddy bear, and wardrobe maintain their positions. Wall inscriptions remain in place, and people appear and disappear, while the weather changes from snow to rain.

The system includes a complex control layer and a lightweight generative layer. The control layer, built around the Latent Stage module, separates the scene, character, camera, lighting, costume, and props in latent space, applying physical constraints for solids, fabrics, liquids, and particles. The rendering layer, LiveStage, uses causal sparse diffusion to sample only the changing areas.

The model has a total of 34 billion parameters but activates about 4 billion per step, which is approximately 11.8 percent. Dawa claims this reduces GPU usage by 40–60 percent and accelerates inference three to five times compared to traditional video diffusion.

On a single consumer NVIDIA GeForce RTX card, the system achieves real-time output at 720p resolution and 25 frames per second. In continuous editing tests, the company reports a 98.2 percent success rate for target editing and a 93.3 percent preservation of human identity. The system utilizes CUDA, TensorRT, and NVENC and operates on local hardware without cloud computing.

Dawa states that its data was gathered from an 84,000 square meter studio space, over 300 real productions, and 180 TB of world state data. The company's Chief Product Officer, Li Zongxin, summarized the approach as follows: 'Video can end, but the world must keep running.' All presented performance metrics are data provided by the company itself.