\n\n

\n\navet is an AV1 encoding service that watches a folder, splits each video at its scene cuts and encodes the chunks in parallel with SVT-AV1. It writes a finished MKV with audio, subtitles and chapters carried over, and ships as a Docker image and as a Linux AppImage with every tool bundled.\n\n

\n\n---\n\n> [!WARNING]\n> This README describes avet 2.0, which is not released yet. The current release is avxs 1.0.0 with the image `ivenos/avxs:latest`, documented in its [README](https://github.com/ivenos/avet/blob/v1.0.0/README.md).\n\n## Features\n\n- Scene-based parallel encoding with [SVT-AV1](https://gitlab.com/AOMediaCodec/SVT-AV1) or [SVT-AV1-HDR](https://github.com/juliobbv-p/svt-av1-hdr)\n- Resumes from the last finished chunk after a restart\n- Target quality: a CRF per chunk from a [CVVDP](https://codeberg.org/Line-fr/Vship) score, with optional [CAMBI](https://github.com/Netflix/vmaf/blob/master/resource/doc/cambi.md) banding limits (GPU required)\n- HDR10, HLG, HDR10+ and Dolby Vision profiles 5, 7 and 8\n- Automatic crop, downscale and keyframe interval\n- Per-track audio and subtitle rules with language filters\n\n## Installation\n\n### Docker\n\n```yaml\nservices:\n avet:\n image: ivenos/avet:latest\n user: \"1000:1000\"\n volumes:\n - ./input:/input\n - ./output:/output\n restart: unless-stopped\n```\n\nWithout `user:` the container runs as root and its files on the host belong to root.\n\n`[target_quality]` needs a GPU, nothing else does:\n\n- Intel or AMD: add `devices: [\"/dev/dri:/dev/dri\"]` and `group_add: [\"render\"]`.\n- NVIDIA: install the [nvidia-container-toolkit](https://github.com/NVIDIA/nvidia-container-toolkit) and add a GPU reservation, or run with `--gpus all`.\n\nThe arm64 image has no hardware Vulkan driver.\n\n### AppImage\n\nDownload the AppImage for your architecture from the [latest release](https://github.com/ivenos/avet/releases/latest) and run it. It watches `input/` and `output/` in the working directory.\n\n## Usage\n\nEvery subfolder of the input directory is a profile: one `encode.toml` and the videos it applies to.\n\n```\ninput/\n├── movies/\n│ ├── encode.toml\n│ └── The Movie (2021).mkv\n├── anime/\n│ ├── encode.toml\n│ └── The Show/\n│ └── Season 1/\n│ └── Episode 01.mkv\n└── processed/ # sources move here once encoded\n\noutput/\n├── The Movie (2021).mkv\n└── The Show/\n └── Season 1/\n └── Episode 01.mkv\n```\n\n- Supported extensions: `mkv`, `mp4`, `mov`, `avi`, `ts`, `m2ts`, `flv`, `webm`, `m4v`. Only the first video track is encoded.\n- Video is encoded as 4:2:0 at 8 or 10 bits. An odd frame size loses its last column or row. Aspect ratio, rotation, the color description, HDR10, HLG and HDR10+ metadata and the offsets between streams are kept.\n- A file whose output already exists is skipped. A file that is still being copied in is picked up once it stops growing.\n- Folders inside a profile are kept in `output/` and `processed/`. Once a folder's last video is encoded, the empty source folder is removed.\n- Output files are named after the source, so two queued files with the same name in the same folder wait until one is renamed.\n- Work in progress lives in `.avet_/` next to the output file. Delete that folder to encode a file from scratch.\n- A failure that can clear on its own, such as a typo in the profile, a timeout or a full disk, is retried on the next scan. Any other failure writes a `.failed` file into that folder, and the video is skipped until you delete it.\n- On `SIGTERM` or `SIGINT` avet finishes the current file, then exits.\n\n## Environment variables\n\n| Variable | Default | Description |\n|---|---|---|\n| `INPUT_DIR` | `./input`, `/input` in the image | Input directory |\n| `OUTPUT_DIR` | `./output`, `/output` in the image | Output directory |\n| `POLL_INTERVAL` | `60` | Seconds between scans |\n| `RUST_LOG` | `info` | Log level, e.g. `debug` |\n\n## Configuration\n\nOnly `encoder` is required. Unknown keys are rejected, so a typo fails the profile instead of turning a feature off.\n\n```toml\nencoder = \"svt-av1\"\n\n[encoder_params]\npreset = 6\ncrf = 28\n\n[avet]\ncrop = true\nkeyint = true\nscale = 1080\nbit_depth = 10\n\n[audio]\nlanguage_whitelist = [\"eng\"]\nmode = \"encode\"\ncodec = \"libopus\"\nbitrate = { stereo = \"192k\", \"5.1\" = \"320k\", \"7.1\" = \"512k\", default = \"192k\" }\n\n[audio.lossless]\ncodec = \"flac\"\noptions = { compression_level = 12 }\n\n[subtitles]\nlanguage_whitelist = [\"eng\", \"jpn\"]\n```\n\n### `encoder`\n\n`svt-av1` or `svt-av1-hdr`. Not needed with `avet.video = \"copy\"`.\n\n### `[encoder_params]`\n\nPassed to the encoder as `--key value`; booleans become `1`/`0`. avet reads two of them itself: `lp` (default `6`) together with free RAM sets how many chunks encode at once, and `crf` is the first probe when `[target_quality]` is set.\n\n### `[target_quality]`\n\nReplaces the fixed `crf`: avet probes each chunk at a few CRF values and encodes at the highest one that still holds `jod`. CVVDP scores in JOD from 0 to 10, where 10 means no visible difference from the source.\n\n```toml\n[target_quality]\njod = 9.5\n```\n\n| Key | Default | Description |\n|---|---|---|\n| `jod` | - | Minimum CVVDP score per chunk, in `(0, 10)` (required) |\n| `min_crf` | `1` | Lowest CRF to try |\n| `max_crf` | `70` | Highest CRF to try (max `70`) |\n| `min_probes` | `2` | Probes before `tolerance` may stop the search |\n| `max_probes` | `7` | Maximum probes per chunk |\n| `tolerance` | `0.05` | Stop once a probe is at most this far above `jod` |\n| `probe_preset` | `13` | Encoder preset for probes |\n| `max_encoded_percent` | `90` | Maximum chunk size as a percent of the source's bytes for that chunk |\n| `max_cambi` | - | Maximum CAMBI of the encode, `>= 0` |\n| `max_cambi_diff` | - | Maximum CAMBI the encode may add on top of its input, `>= 0` |\n\n- `max_encoded_percent` wins over `jod`: a chunk that would grow past it gets a higher CRF, and a warning is logged.\n- CAMBI is `0` without banding; Netflix places slightly annoying banding at around `5`. `max_cambi` counts banding the source already has, `max_cambi_diff` does not. Both apply to the worst 5% of a chunk's frames and are measured without film grain.\n- If no probe holds every limit, the chunk uses the lowest CRF under `max_encoded_percent`.\n\n### `[avet]`\n\n| Key | Default | Description |\n|---|---|---|\n| `video` | `\"encode\"` | `\"copy\"` passes the video through and only processes audio and subtitles |\n| `dv` | `false` | Carry Dolby Vision into the output as AV1 profile 10, converting profile 7 to 8.1 first. Without it, Dolby Vision 7 and 8 keep only their HDR10 base layer and Dolby Vision 5 is refused |\n| `crop` | `false` | Remove black bars |\n| `keyint` | `false` | Keyframe every ~5 s from the frame rate, unless `keyint` is in `[encoder_params]` |\n| `scale` | - | Maximum output height, at least `64`. Taller sources are scaled down with Lanczos |\n| `bit_depth` | - | Encoder input bit depth, `8` or `10`. Unset keeps the source depth, capped at 10 |\n| `keep_temp` | `false` | Keep `.avet_/` after a finished encode |\n\n### `[audio]`\n\n| Key | Default | Description |\n|---|---|---|\n| `mode` | `\"copy\"` | `\"copy\"` or `\"encode\"` |\n| `codec` | - | ffmpeg encoder, e.g. `\"libopus\"`. Required for `\"encode\"` |\n| `bitrate` | - | A single value, or a table keyed by `mono`, `stereo`, `3.0`, `quad`, `5.0`, `5.1`, `6.1`, `7.1` and `default`. Required for lossy codecs |\n| `options` | `{}` | Extra per-track encoder options, e.g. `{ compression_level = 12 }` |\n| `language_whitelist` | `[]` | Keep only these ISO 639-2 languages. Empty keeps all |\n\n- The whitelist matches both spellings (`deu` and `ger`). Tracks without a language or tagged `und` are always kept.\n- Re-encoded tracks get the codec added to their title, e.g. `English 5.1 (Opus)`.\n- Copied tracks Matroska has no codec ID for, such as Blu-ray LPCM, are stored as PCM.\n- Opus gets every channel of a layout it has no mapping for by using the next larger one, e.g. 2.1 as 5.1.\n\n`[audio.lossless]` applies to tracks with a lossless source (`dts` only as DTS-HD MA). `[audio.codec_rules]` applies by source codec as ffprobe names it. Both take the keys above except `language_whitelist`. Unset keys come from `[audio]`, except a non-empty `options`, which replaces it. A matching codec rule wins over `[audio.lossless]`.\n\n```toml\n[audio.codec_rules]\nopus = { mode = \"copy\" }\neac3 = { mode = \"encode\", codec = \"libopus\", bitrate = \"192k\" }\n```\n\n### `[subtitles]`\n\n| Key | Default | Description |\n|---|---|---|\n| `mode` | `\"copy\"` | `\"copy\"` or `\"strip\"` |\n| `language_whitelist` | `[]` | Same rules as for audio |\n\nChapters are always kept. Subtitle tracks Matroska cannot hold, such as TTML, are skipped.\n\n### `[scene_detection]`\n\n| Key | Default | Description |\n|---|---|---|\n| `min_scene_len` | `24` | Minimum chunk length in frames |\n| `extra_split_sec` | `10` | Maximum chunk length in seconds, `0` disables |\n| `extra_split` | `0` | Maximum chunk length in frames (at least `24`), overrides `extra_split_sec` |\n| `speed` | `\"standard\"` | `\"fast\"` trades accuracy for speed |\n| `downscale_height` | - | Detect scenes on a copy scaled to this height (at least `64`) |\n\n## License\n\nCopyright © Iven Schlösser. avet is free software, licensed under the [GNU General Public License v3.0 only](https://github.com/ivenos/avet/blob/main/LICENSE). You may use, modify and redistribute it. Anyone distributing a modified version must release it under the same license and make its source code available.\n\nThe Docker image and the AppImage bundle third-party software, each under its own license:\n\n- [SVT-AV1](https://gitlab.com/AOMediaCodec/SVT-AV1) and [SVT-AV1-HDR](https://github.com/juliobbv-p/svt-av1-hdr): BSD 3-Clause Clear License with the Alliance for Open Media Patent License 1.0\n- [FFmpeg](https://ffmpeg.org): GPL-2.0-or-later and LGPL-2.1-or-later\n- [FFMS2](https://github.com/FFMS/ffms2): MIT as source, GPL as a binary built against FFmpeg\n- [Vship](https://codeberg.org/Line-fr/Vship): MIT NON-AI License\n- [libvmaf](https://github.com/Netflix/vmaf): BSD-2-Clause-Patent\n- [MKVToolNix](https://mkvtoolnix.download): GPL-2.0-only\n\navet is not affiliated with or endorsed by any of them. \"Dolby Vision\" is a trademark of Dolby Laboratories Licensing Corporation, \"HDR10+\" is a trademark of HDR10+ Technologies, LLC.\n","has_starred","hub_user","immutable_tags_settings",{"_264":11,"_265":266},"enabled","rules",[267],".*","is_automated","permissions",{"_271":17,"_272":11,"_273":11},"read","write","admin","user","storage_size",742751496,"isOrg","reponame","tag","sha256:7e929d5c9f184b162a71453c0865e37f9cbd8f6215943c2efb48986fedb3703d","canDeleteTags","isHubSbxKitsEnabled","currentTab","tagAttestations","displaySecurityMetadata","showSbxLogoOverride","routes/_layout.layers.$namespace.$reponame.$tag.(images).($imageUrlDigest)._index",{"_23":24,"_228":-7,"_226":-7,"_227":-7,"_25":289,"_31":280,"_293":-7,"_294":11},{"_27":28,"_29":30,"_31":32,"_33":34,"_35":36,"_37":290,"_63":64,"_65":30,"_66":67,"_68":69,"_70":71,"_72":73,"_74":60,"_75":64,"_76":55,"_77":17},[291,292],{"_40":41,"_31":42,"_43":44,"_45":46,"_47":48,"_49":50,"_51":44,"_52":-5,"_53":34,"_54":55,"_56":-5},{"_40":58,"_31":59,"_43":44,"_45":60,"_47":61,"_49":50,"_51":44,"_52":-5,"_53":62,"_54":55,"_56":-5},"imageWithLayers","hasPermissionToViewRepoSettings","actionData","errors"]

ivenos/avet:dev