Nokia standardizes AI post-filters for H.266/VVC and machine vision efficiency

Nokia’s integration of AI into established standards solves the fragmentation problem of custom neural codecs. By leveraging H.274/VSEI for neural post-filtering, streamers can deploy proprietary AI models that remain backward-compatible with older hardware. This move signals a shift from purely signal-based compression to content-aware, hybrid architectures that bridge the gap between human entertainment and autonomous machine vision. For the ecosystem, this reduces the 'hardware tax' of AI, enabling high-quality 4K and 8K delivery even as global video traffic is projected to double in five years. Watch for the first commercial deployments of standardized VSEI neural filters in premium mobile streaming apps later this year.

The standardization of AI-based media tools reached a critical milestone in early 2025 with the finalization of JPEG AI. Per the Joint Photographic Experts Group (JPEG), this became the first international image coding standard based on end-to-end deep learning. Unlike traditional formats, JPEG AI represents images as latent tensors, providing a 30% efficiency gain over prior methods and offering a template for how video standards like H.267 might eventually move toward fully neural architectures.

In the video sector, the Joint Video Experts Team (JVET) has advanced the Enhanced Compression Model (ECM), which is currently demonstrating roughly 25% bitrate savings over VVC. According to Streaming Learning Center reporting from July 2025, ECM is viewed as the likely foundation for a future H.267 codec. While Nokia is focusing on hybrid pipelines that utilize existing decoders, these parallel tracks indicate the industry is preparing for a transition to 'AI-native' formats by the end of the decade.

Simultaneously, the demand for machine-optimized video is accelerating. Per MPEG documentation from June 2026, the Feature Coding for Machines (FCM) project is finalizing common test conditions to support distributed neural network tasks. Research from Florida Atlantic University in early 2026 highlighted that H.266/VVC provides a significantly better baseline for these machine tasks than H.264, yielding massive bitrate reductions for autonomous systems and smart city infrastructure without sacrificing detection accuracy.