Bilibili Live Streaming Technology for the Spring Festival Gala: Experience Enhancement and Interactive Features

Background Overview

At the end of 2024, Bilibili’s R&D team launched a secret project codenamed CNY to stream the CCTV Spring Festival Gala on New Year's Eve. Anticipating a massive influx of new users, both product and engineering aimed to deliver novel live‑streaming experiences and high‑quality playback. The streaming solution focuses on two major aspects: experience enhancement (picture quality and latency reduction) and gameplay innovation (the "send bullet screen to the gala" feature).

System Introduction

A typical live‑streaming pipeline consists of user push‑stream → source server → transcoding system → downstream CDN distribution. Any jitter at the source is amplified downstream, potentially causing massive QPS spikes when many viewers experience stalls. To mitigate this, Bilibili introduced a cloud‑based broadcast console (cloud director) that performs multi‑source input, automatic switching, and various advanced functions such as seamless switching, dynamic bitrate/resolution adjustment, security frames, custom SEI, super‑resolution enhancement, and external subtitles.

Experience Enhancement

Bilibili launched HDR live streaming, offering higher dynamic range and detail compared to standard SDR. For SDR sources, an AI‑driven SDR‑to‑HDR pipeline runs on the broadcast console. The gala’s source streams arrive in two formats: (1) 4K 50 FPS HDR via SRT and (2) 1080p 25 FPS SDR via RTMP. To serve users unable to play 4K 50 FPS HDR, Bilibili chose an HDR‑to‑SDR conversion using LUT‑based color mapping, integrated into the broadcast console’s binary.

Transcoding quality is evaluated using PSNR (real‑time metric) and calibrated via CRF to meet target PSNR ≈ 44. Performance is measured by average and minimum FPS on various CPU models, with a 10 % safety margin (e.g., targeting 66 FPS for a 60 FPS goal). Latency is measured by timestamping at each pipeline stage (source, transcoding, delivery) and embedding the data in SEI (or AV1 OBU_METADATA) for client‑side reporting.

Bandwidth cost is a critical factor: higher output bitrate increases upload costs for the platform and download costs for users. The system balances quality, bandwidth, and compute by adjusting transcoding parameters and leveraging CDN/P2P strategies.

Gameplay Innovation

The core interactive feature is "send bullet screen to the gala". At a dynamically determined moment during the broadcast, all clients must trigger a synchronized visual effect that aligns with the live gala’s on‑screen bullet screen. To achieve massive simultaneous triggering, Bilibili embeds a custom SEI payload in the video stream containing a UTC timestamp and event‑specific fields. When the event time approaches, clients parse the SEI and trigger the effect precisely.

Operational safeguards include real‑time SEI verification on the broadcast console, batch command dispatch to multiple director instances, and redundant manual/automated execution paths to handle network failures or timing offsets.

Conclusion

The deployment delivered 4K HDR live streaming with ultra‑low latency, supported synchronized interactive bullet‑screen effects for hundreds of millions of viewers, and provided detailed metrics for quality, performance, and bandwidth to guide future decisions.