In Depth: AI Video Is Becoming a Sector to Watch in China, but Don’t Get Out the Popcorn Yet

Artificial intelligence (AI)-driven video generation tools have rapidly become a focal point for investment among major Chinese tech firms seeking to diversify their revenue streams. Since OpenAI’s groundbreaking release of Sora, a text-to-video model, in February 2024, Chinese companies have hastened to roll out competitive offerings, enabling the creation of short films and video series through AI. Notably, Kuaishou Technology and ByteDance, two leading players, have launched their own AI video tools and introduced subscription models for their use, although the technology is still limited in its capability to transform the broader filmmaking industry or related sectors [para. 1][para. 2][para. 3][para. 4].

Kuaishou, diverging from the industry trend of chatbot development, has focused on large language models (LLMs) primarily for its short-video app and advertising businesses [para. 5]. Senior vice president Gai Kun stated that generative pre-training transformers face hurdles in attaining robust reasoning powers, rendering a move toward artificial general intelligence challenging [para. 5][para. 6]. Moreover, Kuaishou’s LLMs have not met commercial expectations, prompting the company to identify AI video generation as a new avenue for growth [para. 6]. To that end, Kuaishou has invested heavily in foundational video generation models, constructing a large computing cluster over two years to train models encompassing trillions of parameters. In August 2024, the company highlighted these investments and, by the following June, launched Kling AI, which as of its 2.1 release could generate five-second, 1080p videos [para. 7]. Kling AI generated 150 million yuan (~$20.8 million) in revenue in Q1 2025, becoming a significant new revenue stream [para. 8].

ByteDance has similarly prioritized video generation, restructuring leadership so that former Douyin CEO Zhang Nan could spearhead AI projects within its video-editing division [para. 9]. Zhang’s team had increased computational resources to develop Jimeng AI, inspired by OpenAI’s Dall-E 2, which creates images from textual prompts [para. 10]. In September 2024, ByteDance launched additional video generation models for Jimeng AI, with five now integrated and managed by the company’s Seed department, focused on both internal and external applications [para. 11]. In March 2025, Jimeng AI also incorporated DeepSeek’s R1 reasoning model to enhance users’ text prompt detail [para. 12]. As of March 2025, Jimeng AI had 8.93 million monthly active users, while Kling AI counted 1.8 million, according to QuestMobile [para. 13].

Video creators report that different AI tools excel in different areas: Kling AI is better for precise character nuances, while Jimeng AI excels in scene continuity and character movement [para. 14]. The rise of these tools has given way to a creator economy, with tiered subscription systems launched soon after the introduction of Kling AI and Jimeng AI, offering various credit packages [para. 15][para. 17]. Kuaishou introduced initiatives to support creators in landing contracts, while ByteDance ran campaigns to distribute Jimeng credits and foster micro drama creation, resulting in new opportunities in e-commerce-linked content [para. 18][para. 19][para. 20].

Despite these advances, current AI video generators remain unable to produce seamless, feature-length films automatically [para. 22]. Filmmakers like Wei Lian highlight the necessity of using multiple tools—such as Jimeng AI, Kling AI, PixVerse, and Pika AI—due to each tool's unique strengths, indicating no single solution dominates yet [para. 23][para. 24]. Future improvements are expected as multimodal models, which integrate text, images, audio, and video, become more sophisticated and capable of generating more realistic, complex videos [para. 25].