FFmpeg 7.0

Original Article ↗ Hacker News Discussion ↗

Notable FFmpeg 7.0 Features

  • Big feature list: parallelized ffmpeg CLI pipeline, new filters (QR encode, tilt/shift, fsync, dnn with libtorch), new demuxers/muxers (DVD-Video via libdvdnav/libdvdread, IAMF, RCWT, AEA), HW decoding via D3D12VA, VVC decoder (experimental), optimizations for RISC‑V, LoongArch, AArch64.
  • New metadata and HDR features: IAMF/ambisonics, HEIF/AVIF stills, Dolby Vision profile 10 in AV1, HDR10 metadata passthrough.
  • Tooling changes: ffprobe enhancements, removal of deprecated CLI options, default QSV bitrate mode changed, new -bsf behavior, -@/opt file-based options.

CLI Multithreading and Loopback Decoders

  • Clarified that multithreading existed in codecs already; 7.0 adds parallelism across demuxing/decoding/filtering/encoding/muxing in the CLI, potentially improving throughput, especially with multiple outputs.
  • Loopback decoders are introduced; unclear to some how they help with dynamic tiling workflows versus simply cloning filter outputs. Some confusion remains about their practical use.

New Codecs and Video Standards

  • VVC decoder exists but is flagged experimental and currently slow; external VVdec is reported to be 2–3× faster on AVX2/SSE4 hardware, but integration patches only work up to 6.1.1.
  • Debate over usefulness of more codecs: VVC vs AV1 vs MPEG‑5 EVC, with claims VVC offers ~20–30% bitrate reduction vs AV1 and is already being chosen for some broadcast standards.
  • Interest in IAMF/ambisonics but uncertainty about tooling and how to experiment with it.

Machine Learning Filters

  • ffmpeg has ML-based filters with TensorFlow, Torch, and OpenVINO backends; some see this as fragmented and ask for a more coherent strategy and model management (e.g., official model repositories).
  • Desire for built‑in workflows for super‑resolution and automatic subtitles (e.g., Whisper), but no clear roadmap mentioned.

CLI Complexity, GUIs, and WASM

  • Strong split: some argue CLI + docs/examples are fine for power users; others find the CLI byzantine and want simpler “90% use case” guidance or wizards.
  • No GUI fully covers ffmpeg’s capabilities; various tools (HandBrake, LosslessCut, Kdenlive, XMedia Recode, etc.) are mentioned as partial frontends.
  • Debate whether tools like HandBrake “are ffmpeg” (via libav* libraries) versus being limited GUIs that expose only a subset of ffmpeg’s power.
  • Proposal for an in-browser ffmpeg (WASM) with a friendly UI to replace shady conversion sites; others question browser performance/complexity and suggest native apps instead.

LLMs as ffmpeg Assistants

  • Many report LLMs are unusually good at generating ffmpeg commands and explanations, making the tool far more accessible for occasional users.
  • Others warn about hallucinated options and subtle bugs; see LLMs as useful starting points only when the user can verify results.

Packaging, Dependencies, and Toolchains

  • Some package ecosystems (winget, MacPorts, Homebrew) lag behind 7.0 by hours to much longer; discussion around open contribution to manifests and frustration with specific package managers.
  • Builds via vcpkg are temporarily affected by GitHub’s blocking of liblzma/xz downloads due to the recent backdoor, prompting debate over security vs access to compromised history.
  • Move to requiring C11 (and signaling a future C17 requirement) draws criticism from at least one commenter who sees frequent language standard bumps as unnecessary and risky.

Praise, History, and Use Cases

  • Widespread admiration for ffmpeg as a foundational FOSS “wonder,” powering everything from kids converting videos for old iPods/PS2s to massive transcoding pipelines, video mosaics, and scripted workflows.
  • Nostalgic recollections of early 2000s codec chaos (DivX, Xvid, codec packs) contrasted with today’s relative simplicity thanks to ffmpeg and modern players.