-- At the end of 2025, the Institute of Artificial Intelligence of China Telecom (TeleAI), announced a major breakthrough: Generative Video Compression (GVC). Leveraging the visual priors of generative AI, this technology extracts and transmits compact perceptual features and motion dynamics. At the receiving end, it relies on the powerful generative capabilities of large models to real-time ‘infer’ and reconstruct high-fidelity video with realistic textures and restored details. GVC achieves an unprecedented compression rate as low as 0.02% while preserving visual clarity and exceptional temporal smoothness. It enables seamless, zero-lag video calls, live streaming, and task-oriented video understanding even in weak-signal or bandwidth-constrained environments, such as remote rural areas and maritime zones.
Classical communication theory, rooted in the Shannon-Weaver model introduced in the 1940s, conceptualizes communication across three levels: Level A addresses technical data transmission; Level B considers semantic meaning; and Level C focuses on effectiveness, or task-oriented communication. For decades, video communication has primarily focused on optimizing Level A or B to maintain pixel fidelity. However, traditional methods often fail in weak networks, resulting in buffering, freezing, or pixelation when bandwidth is insufficient. GVC bridges this gap by shifting the focus to Level C, prioritizing the effectiveness of the information delivered.
“In the era of AI, the essence of communication has evolved from data transmission to intelligence distribution,” said Prof. Dr. Xuelong Li, CTO and Chief Scientist of China Telecom and Director of TeleAI. “GVC ushers in a new phase of video communication, shifting the core logic from ‘pixel transportation’ to ‘semantic generation’. By trading computation for compression rate, it paves a viable path for an advanced video communication paradigm that is effective, efficient, scalable and practical.”
Building on the AI Flow framework pioneered by Prof. Dr. Xuelong Li in 2024, the research team of TeleAI introduced the Law of Information Capacity to measure the knowledge density and intelligence capability of models via data compression. At the heart of GVC is the core philosophy of "trading computation for bandwidth." By utilizing model distillation, quantization, and powerful generative priors, GVC allows the receiver to synthesize missing details rather than transmit them. This dramatically reduces communication overhead, ensuring that even consumer-grade GPUs can perform low-latency inference. Consequently, the video stream remains visually fluid and coherent, completely eliminating the "stuttering" or "frame skipping" typical of low-bandwidth connections.
Unlike traditional codecs like HEVC that struggle to preserve every pixel, GVC adopts a "Perception + Task-Oriented" mode, transmitting only the "perceptual guidance" features for human vision and "semantically effective" representations for machine understanding. In standard benchmark tests (MCL-JCV), GVC achieves an extreme compression rate of 0.005 bpp (~0.02%) while delivering visually compelling reconstructions. Even under such extreme compression, the generated video maintains smooth motion and consistent semantics, ensuring a high-quality experience without the artifacts usually seen in low-bitrate streams.
In conclusion, GVC, developed by the Institute of Artificial Intelligence of China Telecom (TeleAI), significantly reduces transmission bandwidth requirements while maintaining high perceptual quality. It is particularly well-suited for scenarios demanding high efficiency and uninterrupted fluidity, such as maritime communications via narrowband satellites, emergency rescue operations, remote video surveillance under narrowband mobile networks, and edge computing on in-vehicle or wearable devices. By transmitting only the core information necessary for perception and decision-making, GVC not only defines a new technical paradigm but also redefines the essence of communication: guaranteeing a smooth, intelligent, and adaptive connection regardless of network conditions.
To learn more about the Institute of Artificial Intelligence of China Telecom (TeleAI), click here: https://www.teleai.com.cn/product/AboutTeleAI
To read the technical report of Generative Video Compression (GVC), click here:
https://www.arxiv.org/abs/2512.24300
Contact Info:
Name: Ziyao Tang
Email: Send Email
Organization: TeleAI
Website: https://www.teleai.com.cn/product/AboutTeleAI
Release ID: 89182164
In the event of encountering any errors, concerns, or inconsistencies within the content shared in this press release, we kindly request that you immediately contact us at error@releasecontact.com (it is important to note that this email is the authorized channel for such matters, sending multiple emails to multiple addresses does not necessarily help expedite your request). Our dedicated team will be readily accessible to address your feedback within 8 hours and take appropriate measures to rectify any identified issues or facilitate press release takedowns. Ensuring accuracy and reliability are central to our commitment.
