Media filtering is usually implemented as a feature inside a particular player. OpenFilter Lab asks what changes when the filter becomes a portable artifact the viewer can inspect, edit, share, and carry between compatible tools.
A proposed .of.json sidecar format describing how media was matched, timestamped events, categories, severity, actions, provenance, and review state.
The aim. Let people own transparent media filters that can move between tools instead of being locked inside a platform.
01The problem behind the project
Filtering tools usually hide their decisions or trap them in one application. A portable sidecar makes annotations reviewable, editable, and locally enforceable.
The proposed .of.json sidecar records how a media file was matched, which timestamped events were identified, what categories and severity were assigned, what action is requested, and whether the annotation came from a person or an unreviewed model draft.
Viewers, parents, accessibility users, media-tool developers, and communities with specific content preferences may benefit. Creators and rights holders are affected by derivative annotations.
02How it took shape
A schema, CLI workflow using ffprobe and ffmpeg, optional classifier-generated drafts requiring review, and a local player that can consume mute and skip events.
The laboratory workflow uses ffprobe and ffmpeg, an optional classifier, human review, and a local player capable of consuming mute and skip events. The classifier is intentionally not the final authority; its useful output is a draft that preserves provenance and can be corrected.
Josiah developed the portable-filter idea, local-first values, human-review rule, and product direction, then directed agent-assisted prototyping.
A functioning laboratory repository and player workflow exist, but there is no broad ecosystem adoption or mature classifier.
03What the project means now
The larger idea is user-owned media policy without bypassing access controls. Interoperability could make household preferences durable, but copyright, DRM, matching reliability, and annotation rights define hard boundaries for any public ecosystem.
Copyright, DRM, platform terms, matching reliability, and annotation quality constrain deployment. The project does not bypass access controls.
The trustworthy output of an AI media classifier is a reviewable draft annotation, not an invisible enforcement decision.
Stabilize the sidecar specification, publish sample media-safe fixtures, and test interoperability between two independent consumers.