Show HN: Ez FFmpeg – Video editing in plain English

Original Article ↗ Hacker News Discussion ↗

LLMs as ffmpeg frontends

  • Many use chatbots specifically to generate ffmpeg commands (including complex effects like “bounce” forward/backward loops), iterating until it works and then saving the command.
  • People emphasize LLMs as “text → formal directive” translators (CLI, SQL, legalese), with the user still inspecting and learning from the output.
  • Others report LLM hallucinations on ffmpeg edge cases and stress that understanding codecs/containers is still often necessary.

Learning vs offloading complexity

  • Some argue it’s not worth learning ffmpeg if used once or twice a year; they just want a one-off solution.
  • Others counter that you don’t memorize options, you learn concepts and where to look in the docs; LLMs can’t yet be trusted like well‑vetted libraries or distro packages.
  • Debate over “trust”: library code is implicitly tested in real-world use; LLM output must be individually verified.

Value and limits of plain-English wrappers

  • Supporters like ezff for covering common patterns (“90% use cases”) and avoiding manual syntax.
  • Critics say oversimplified wrappers that always re‑encode (e.g., mkv→mp4, extracting audio as MP3) hide important distinctions like remux vs transcode and quality settings, reinforcing misconceptions about video.
  • Concern about the “cliff”: once a need falls outside the 20 patterns, users are stuck without having learned ffmpeg itself; some suggest falling back to an LLM in those cases.

Implementation concerns (Node/npm)

  • Multiple commenters dislike a CLI that depends on Node/npm due to bloat and supply‑chain risk, preferring a single static binary (Rust/Go/C) or distro packages.
  • The GitHub link on npm returns 404, which some find off‑putting; people resort to the “code” tab on npm.

Alternative UX ideas and tools

  • Suggestions: tab completion, TUI/interactive builders that explain each flag, “typed” CLIs with dropdown-like choices, or generalized helpers like helpme and llmwrap.
  • Some rely on GUI tools (e.g., Lossless Cut, ScreenFlow) or their own wrapper scripts/aliases instead.

Video/GIF and codec tangents

  • Discussion on macOS failing to open some MP4s due to codecs; advice to re‑encode to H.264+AAC or use players like VLC/mpv/IINA.
  • Multiple detailed ffmpeg snippets for high‑quality GIFs (palettegen, paletteuse, dithering), mentions of gifski and palette‑per‑frame.
  • One nitpick about ambiguous phrasing like “slow down by 2x” in the English interface.