HybridPrompt: Bridging Generative Priors and Traditional Codecs for Mobile Streaming Conference: Workshop on Network and Operating System Support for Digital Audio and Video; April 4–8, 2026; Hong Kong, Hong Kong Workshop on Network and Operating System Support for Digital Audio and Video (NOSSDAV ’
Liming Liu, Jiangkai Wu, Haoyang Wang, Peiheng Wang, Zongming Guo, Xinggong Zhang · 2026
Read HybridPrompt: Bridging Generative Priors and Traditional Codecs for Mobile Streaming Conference: Workshop on Network and Operating System Support for Digital Audio and Video; April 4–8, 2026; Hong Kong, Hong Kong Workshop on Network and Operating System Support for Digital Audio and Video (NOSSDAV ’ by Liming Liu, Jiangkai Wu, Haoyang Wang, Peiheng Wang, Zongming Guo, Xinggong Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-17120.txt.
Description
Liming Liu led a six-author team chasing real-time, high-quality video streaming on phones. Traditional codecs decode fast but degrade badly at low bandwidth, while generative neural codecs look far better yet demand heavy frame-by-frame generation that phones cannot sustain. The paper asks whether both strengths can combine. Its answer is HybridPrompt, the first generative video system to decode full high-definition video in real time on a commercial smartphone.
Keyframes go through a generative model while traditional codecs handle the rest. A key insight makes the pairing work: the traditional decoding process is differentiable. Later frames can thus supervise keyframe generation end to end. The generative model thus crafts keyframes that are both good-looking and ideal references for the codec. A two-stage generation strategy rounds out a system far faster than pure neural baselines and visibly better than traditional codecs at low bitrates.
Catalog record
| CREATOR | Liming Liu, Jiangkai Wu, Haoyang Wang, Peiheng Wang, Zongming Guo, Xinggong Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-17120