#fountaincode — Public Fediverse posts

----------------

🎥 Video
===================

Executive summary: A technical demonstration walks through converting arbitrary files into video containers for storage on YouTube. The project documents practical constraints (YouTube file/length limits, metadata stripping, and aggressive transcoding) and presents a workflow combining chunking, integrity checks, and forward error correction to enable file reconstruction after upload.

Technical details:
• Encapsulation: The workflow targets standard video containers and uses video and audio tracks as the durable carriers because YouTube strips most metadata and can reject subtitle payloads.
• Integrity checks: Uses multiple CRC flavors to detect corrupted chunks prior to reconstruction.
• Forward error correction: Implements Wirehair (an O(N) fountain code) to create redundant symbols so that the original file can be recovered despite dropped or heavily altered chunks during YouTube transcoding.
• Encoding channel: Embeds payload bits into transform-domain coefficients — specifically leveraging the Discrete Cosine Transform (DCT) used by common codecs — to hide data within compressed frames while balancing capacity and survivability.

Implementation concepts:
• Chunking strategy: Files are split into chunks sized to fit per-video capacity limits (YouTube supports up to 256 GB or 12 hours), then encoded into frames or audio payloads with added FEC symbols.
• Hybrid error-proof algorithm: Combines CRC validation for corruption detection with fountain-code-based redundancy for recovery of missing symbols.
• Codec selection: Emphasizes that codec choice and compression aggressiveness materially affect recoverability; lower-loss codecs and control of quantization on DCT coefficients increase success rates.

Use cases and limitations:
• Practical use cases include long-term archival of very large files and covert transport where traditional storage is unavailable. The approach is constrained by platform policy, upload limits, potential content removal, and the non-deterministic nature of platform transcoding pipelines.

Detection and considerations:
• Detection vectors are platform-specific; artifacts include atypical frame-level entropy patterns and persistent non-media payloads in transform coefficients. The talk notes that subtitles/metadata are unreliable for storage because of sanitization.

References and tooling:
• The presentation references the Wirehair fountain codec and recommends studying CRC variants and video compression internals. Visualizations were created with Manim and DaVinci Resolve.

🔹 wirehair #fountaincode #crc #dct #tool

🔗 Source: https://www.youtube.com/watch?v=l03Os5uwWmk