Get Started
Home
Topics
Search
Library
Research questionHow can ultra-low-bitrate video compression preserve perceptual quality and temporal consistency while decoding in real time?At very low bitrates, conventional neural codecs can blur detail, while generative codecs may decode slowly or produce temporally inconsistent frames. The central difficulty is meeting fidelity and consistency requirements without making playback too slow.
AI
Computer Vision
Diffusion Models
Image & Video Processing
Inference Optimization
Technology
Video Generation
Latest papersRecent research connected to this question, newest first.VoRTeC: Taming Foundation Flow for One-step Real time Video CompressionThe source addresses flow-based generative video compression with one-step decoding, without access to the flow-matching network’s parameters or gradients. It reports results at 720p and 480p, including 13 FPS and 32 FPS decoding speeds respectively, along with lower bit consumption than prior diffusion-based approaches; these are source-reported results rather than general guarantees.research paper · Sep 3, 2026
Related questions
How can neural video representation decoders reduce compute and memory use on resource-constrained devices without hurting reconstruction quality?How can learned image codecs provide progressively refinable reconstructions from a single bitstream?How can implicit-neural video codecs decouple bitrate scalability from decoding complexity in one embedded bitstream?How can video diffusion models be quantized for efficient deployment without losing fine visual detail?