jelly_coder scans a folder for supported videos (.mkv, .mp4, .wmv, .mwv, .avi, .webm) and re-encodes them with size-aware defaults. It keeps subtitles/metadata, flips containers when MP4 constraints are violated, and preserves a mirrored directory tree when you choose not to overwrite in place.
Key capabilities:
- Automatic encoder selection with hardware-first preference (NVENC → QSV → AMF → x264) and explicit backend override flags.
- Height presets (
auto,1080p,720p,480p,360p) with bitrate scaling so smaller outputs also shrink file size. - Smart audio handling: copies compatible streams, forces stereo when sources are mono or pseudo-mono (e.g., WMV files with only one active channel), and warns about downmixing.
- Hardware fallbacks: if a chosen hardware encoder fails to create a session, the tool transparently repeats the job with
libx264so the batch continues.
- Windows PowerShell 5.1 (default shell for the repo scripts).
- Python 3.11+ with virtual environment support.
- ffmpeg and ffprobe in
PATH.- Hardware backends require vendor drivers and an ffmpeg build with NVENC/QSV/AMF enabled.
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install --upgrade pip
pip install -e .# Show CLI help
python -m jelly_coder --help
# Reduce a folder, mirror outputs under ./output/<dir>, auto-select backend
python -m jelly_coder --input D:\media --quality 720p
# Force Intel QSV, enable overwrite, run at debug verbosity
python -m jelly_coder --input D:\media --encoder-backend qsv --overwrite --log-level debug
# Legacy wrapper remains available
python encode_videos.py --input D:\media--input PATH: Directory scanned recursively for supported video extensions.--encoder-backend BACKEND:auto(default),nvenc,x264,qsv, oramf.--preferred-codec CODEC: Hinth264orhevc; respected when the backend supports it.--quality PRESET: Downscale preset (auto,1080p,720p,480p,360p).--workers N: Concurrent encodes (default 1; hardware encoders generally behave best at 1).--overwrite: Replace sources in place. When omitted, outputs land in./output/<input-folder>.--log-level LEVEL:info(default),debug,warning, etc.
- MP4 targets convert SubRip/ASS to
mov_textautomatically; incompatible streams trigger a fallback to MKV. - Attached artwork, metadata, and subtitle tracks are propagated.
- When the output is larger than the input, a warning is emitted so you can review or delete the file.
- Audio streams are copied when safe; otherwise they are re-encoded to AAC 192k stereo, duplicating the dominant channel if the source is effectively mono.
- Auto: Queries ffmpeg encoders and picks the best available hardware backend (NVENC → QSV → AMF → x264).
- NVENC/QSV/AMF: Uses vendor hardware. Hardware decode fallbacks are attempted before giving up.
- x264: Software-only; always available.
- If a hardware backend exhausts its attempts (e.g.,
Error creating a MFX session: -9on QSV), the utility logs a warning and re-runs the encode withlibx264automatically.
- ffprobe insights drive copy-or-transcode decisions.
- WMV inputs and other stereo tracks with severe channel imbalance trigger a sampling step; if only one channel has meaningful signal, the tool duplicates it so the result plays on both speakers.
- Mono and surround sources are converted to stereo with informative log messages about the change.
- Run
python -m pytestandflake8before committing; coverage is enforced at 100%. python -m jelly_coder --helpverifies CLI wiring after refactors.- VS Code launchers in
.vscode/launch.jsonprovide ready-to-run debug sessions.
- Validate ffmpeg exposes the expected encoders:
ffmpeg -hide_banner -encoders | Select-String nvenc. - Hardware backends may need up-to-date GPU/Media drivers; keep Windows and vendor packages current.
- Delete partial outputs in
output/if you want to re-run without--overwrite.