| output-spec.md |
Configuring a transcode — the complete OutputSpec guide: every builder method, enum, and field, plus how to run a job. |
| filters/ |
Video filters — a page per filter (crop, pad, flip, rotate, grayscale, overlay, brightness/contrast/saturation/invert, the denoise family, parameterized nlmeans, and the temporal hqdn3d), the string + structured-object forms, and per-surface usage. |
| lossless-audio.md |
Lossless audio — FLAC and ALAC: why a web-first engine carries them, the settings, container rules and codec strings, channel order, browser support, and how they were verified. |
| audio-filters.md |
Audio filters — channelmap (remap / reorder / select channels), the channel + layout vocabulary, and how the input layout is resolved. |
| batch.md |
Batch manifest DSL — convert many files from one YAML/JSON file (rivet batch): the manifest shape, every key, glob inputs, output rules, and examples. |
| cli.md |
rivet CLI reference — every subcommand, flag, and environment variable, with examples. |
| hooks.md |
Hooks: a specific kind for each point of a job (source bytes, probe, decoded frames, encoder frames, stills, artifacts, completed, failed). Verdicts, blocking vs background, fail open or closed, per-job reports, the built-in digest and perceptual-fingerprint hooks, and the HTTP API's /v1/hooks. |
| hooks-cookbook.md |
Hook cookbook: sixteen worked recipes (size and container gates, source-material hashing, probe limits, blank-frame detection, your own hashing library, filter drift, background forwarding, output caps and manifests, metrics, rejected-job reports, policies, the HTTP API, unit-testing hooks), all compiled in examples/hook_cookbook.rs. |
| inference.md |
Running models on a job's pictures: the guide to inference on hooks. Which point and which frames, getting pixels into a model (planar_f32_letterboxed, normalisation, letterbox coordinates, HDR sources), choosing and running a runtime (ONNX Runtime and its CUDA / DirectML / OpenVINO providers, candle, tract, a model server), warm-up, session pools and CUDA graphs, annotations and gating verdicts, getting results out (report, live forwarding, HTTP API), measured performance, deployment (runtime libraries, containers, Intel compute runtime, Resizable BAR), testing, and a checklist. |
| hooks-yolo.md |
YOLO object detection with hooks: a YOLO detector (ONNX, through ONNX Runtime) as a decoded-frame and still hook. Getting a model and the runtime, every output layout (v5 to 26), what each picture goes through, reading the boxes from the report, rejecting jobs on what they show, GPUs, throughput, and other runtimes. The code is the examples/yolo crate. |
| api.md |
HTTP transcode API (rivet serve) — endpoints, request bodies, the job lifecycle, and the OpenAPI / Swagger / Redoc docs. |
| ../bench/ |
Quality bench — the VMAF/SSIM harness: a reproducible corpus, a scorer that upscales each rung to source and scores a mid-clip window, and run-ladder.sh to go from a clip and any set of rivet transcode flags to a scored ladder. A ladder change is not a result until it has been scored. |