On March 4, 2026, Huoshan Engine, a subsidiary of ByteDance, officially announced the pricing for its Seedance2.0 video generation model, marking the first clear and actionable commercial billing benchmark in China's AI video industry.

Multiple Doubts

The official pricing adopts a differentiated token-based billing model: 28 yuan per million tokens for video input (video editing) and 46 yuan per million tokens for non-video input (pure text-to-video generation). The core difference lies in computational power consumption. Video editing optimizes based on existing materials, requiring lower computational power.

Pure video generation builds scenes, timing, and physical logic from scratch, leading to exponential growth in computational power consumption and, consequently, higher pricing.

Based on real-world testing data, generating a 15-second video requires 308,880 tokens. Using the pure generation model, the cost per video is approximately 15 yuan, translating to 1 yuan per second. This marks the industry's official entry into the "yuan-per-second era."

As the core video capability of ByteDance's Doubao large model, Seedance2.0 emphasizes character consistency, long-term temporal stability, and multi-shot storytelling, earning it the nickname "director-level AI." Currently, it is only available to leading companies in Manhua Drama (manhua dramas), short dramas, and marketing, with no immediate plans for full API access. Its cautious commercialization approach has raised multiple doubts in the market regarding its pricing logic, profit model, and strategic intentions.

Huoshan Engine, a fully-owned enterprise-level cloud and AI service platform under ByteDance, carries the core mission of technology spillover and B-side monetization for the group. The likelihood of an independent IPO in the short term is extremely low.

On one hand, it is deeply integrated with ecosystems such as Douyin, Jianying, and TikTok, with highly internalized loops of computational power, models, data, and scenarios. A spin-off would disrupt synergistic efficiency and data closed loop (closed loops).

On the other hand, ByteDance is in a period of all-encompassing AI investment, with high capital expenditures required for large models, multimodal technologies, and computational infrastructure. The conflict between listed company financial reporting pressures and long-term strategy has led the market to view it as ByteDance's core non-listed growth engine.

The company is characterized by strong technology and scenario integration but weak independent profitability. Internally, it supports the content ecosystem upgrade of the entire group. Externally, it provides MaaS services through Huoshan Fangzhou, with the Seedance series as its flagship product.

At the product level, Seedance2.0 addresses industry pain points such as character drift, physical distortions, and audio-visual asynchrony. It supports multimodal reference inputs, significantly improving video usability. However, its shortcomings are also evident: it focuses solely on 15-second short videos, lacks long-form video capabilities, and has closed APIs limited to internal ecosystem use. Its technological openness and commercial coverage are far inferior to peers.

Key Leap

From an industry landscape and market value perspective, AI video generation is on the brink of explosive growth. Goldman Sachs predicts that the global market size will exceed $29 billion by 2030. Domestically, sectors such as short dramas, e-commerce marketing, and animation asset generation are experiencing rapid penetration. By 2025, nearly 40% of leading short dramas will adopt AI-generated technology, marking a critical leap from technical validation to industrial deployment.

The pricing implementation of Seedance2.0 represents a key step for ByteDance in completing the closed loop of "large models, multimodality, video generation, and commercialization." Leveraging Douyin's ecological traffic and data advantages, it aims to gain a foothold in three trillion-yuan markets: short drama industrialization, enterprise marketing materials, and digital content customization.

However, industry risks remain acute: First, the rigid pressure of computational power costs. Video generation consumes massive computational resources, with GPU supply and pricing directly impacting gross margins. Whether the 1-yuan-per-second pricing can decrease with scale remains uncertain.

Second, intensifying competition. Domestic players like Kling, Wanxiang, and MiniMax are iterating rapidly, while overseas players like Sora and Veo possess deep technological reserves, exacerbating risks of homogenization and price wars.

Third, regulatory and copyright risks. Issues such as deepfakes, portrait rights, and training data copyright remain unresolved, with stricter regulations increasing compliance costs.

Fourth, commercialization falling short of expectations. B-side clients have extremely high demands for effectiveness, stability, and workflow integration. Current models still cannot fully replace professional production.

Deep-Seated Game

In the long run, the pricing of Seedance2.0 marks a symbolic event in the transition of AI video from "showcasing technology" to "calculating costs," reflecting the deep-seated game between the content industry and AI technology integration.

Industry trends are clear: Video is the next-generation content gateway, and AI video is the digital content infrastructure. It will fundamentally reconstruct production processes in short dramas, advertising, and film and television, achieving cost reduction, efficiency gains, and industrial-scale production.

For Huoshan Engine and ByteDance, reliance on ByteDance's ecosystem is both an advantage and a shackle. Their ability to expand independently remains unproven. Technological leadership is merely an entry ticket; cost control, scenario deployment, and ecological construction are the keys to victory. The 1-yuan-per-second pricing, while establishing a commercial benchmark, exposes core issues such as high computational costs, narrow profit margins, and low openness.

If they cannot rapidly reduce computational costs, open APIs to expand the ecosystem, and break through long-form video generation bottlenecks, they may fall passive in industry price wars and technological iterations, despite their first-mover advantage.

For the market and investors, Seedance2.0's "1 yuan per second" represents both hope for AI video industrialization and a cost-profit test hanging over the industry. The true commercialization inflection point remains far off.