Jev 101
Start Guides
Search
中文 EN
Media heat 2 · Cool★ 0heat incomplete (no X)

FrameFind

A local Python app that finds moments in your videos from a plain-language query: FFmpeg samples every tenth frame, Jev checks each sampled frame against the query, and matches above a fixed cutoff become playable, exportable snippets.

Repo: Lyadalachanchu/framefind ↗

What it is

Lets you search footage the way you would describe it.

What it does: Scores sampled frames against your query and returns timed snippets.

Why it belongs in the showcase: A clear, cost-guarded pattern for image judgments over video.

Caveats: Visual only; preview feature.

Evidence

Public MIT repo with CI on Linux and macOS, created 2026-10-20. Default model jev-1.14.0 with 8 concurrent requests; the cutoff is fixed at 95% relevance (a normalized score, not a probability of correctness). Repeat queries reuse saved scores. A $15 local spending guard is on by default. Videos, frames and the SQLite database stay local; sampled images and the query go to TypeSafe. Tests use generated media and mocked responses.

Caveats

  • Searches visible content only, not speech or audio; sampling can miss brief events
  • Image input is a TypeSafe preview feature
  • Only import videos you have the right to use

Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.