lightflow-video-auto-edit

# LightFlow Video Auto Edit Use `lightflow.video_auto_edit` only after `lightflow.video_highlights` has returned its HMAC-verified `clips` output. Pass that output unchanged as `sources`. The workflow renders a normalized MP4 from explicit verified ranges; it never creates segments from silence, scene changes, or filename metadata. ## Runtime The published workflow crate contains a Rust-native `runner.v1` binary. No repository-relative script or global command runner is required. FFmpeg and FFprobe are the only host media dependencies and missing tools fail closed. ## Contract - Required `sources`: JSON array of clip objects with VideoScore provenance. - Required `brief`: editing goal used for deterministic ranking. - Required `output_path`: destination MP4. - Optional `style`: defaults to `clean social edit`. - Optional `constraints`: supports `aspect_ratio`, `max_duration_seconds`, `fps`, `width`, and `height`. - Outputs: `edit_plan`, `video`, `video_path`, and `summary`. Set `LIGHTFLOW_VIDEOSCORE_EVIDENCE_KEY` in the runtime environment for both the highlight and auto-edit workflows. Each clip requires `path` and a `highlight` object. Preserve an explicit `start` / `end` range when supplied; the highlight's `source_path` must resolve to the same source, its `start_seconds` / `end_seconds` must match that range within 0.001 seconds, its numeric `score` must be 1 through 4, and both `model` and `reason` must be non-empty. Require `workflow:"lightflow.video_highlights"`, `model:"TIGER-Lab/VideoScore-v1.1"`, and an `evidence` lowercase-hex HMAC tag generated by the highlight workflow. String-only sources and missing or mismatched evidence are rejected. ## CLI Usage From this project root: ```bash lfw run lightflow.video_auto_edit \ --input sources='[{"id":"intro","path":"media/intro.mp4","start":0,"end":8,"highlight":{"workflow":"lightflow.video_highlights","source_path":"media/intro.mp4","start_seconds":0,"end_seconds":8,"score":3.7,"model":"TIGER-Lab/VideoScore-v1.1","reason":"Strong full-vehicle opening.","evidence":"<generated-by-lightflow.video_highlights>"}}]' \ --input brief='"Create a concise product recap with a strong hook."' \ --input output_path='"output/recap.mp4"' \ --input constraints='{"aspect_ratio":"9:16","max_duration_seconds":30}' ``` When `lfw` is not installed, replace `lfw` with `cargo run --manifest-path ../../Cargo.toml --bin lfw --`. ## API Usage Start `lfw serve`, then use the shared HTTP workflow contract: ```bash curl -sS -X POST http://127.0.0.1:5174/workflows/lightflow.video_auto_edit/run \ -H 'content-type: application/json' \ -d '{"inputs":{"sources":[{"path":"media/intro.mp4","start":0,"end":8,"highlight":{"workflow":"lightflow.video_highlights","source_path":"media/intro.mp4","start_seconds":0,"end_seconds":8,"score":3.7,"model":"TIGER-Lab/VideoScore-v1.1","reason":"Strong full-vehicle opening.","evidence":"<generated-by-lightflow.video_highlights>"}}],"brief":"Create a concise product recap with a strong hook.","output_path":"output/recap.mp4","constraints":{"aspect_ratio":"9:16","max_duration_seconds":30}}}' ``` Run the crate's Rust tests and Clippy after changing the runner, planner, edit-plan schema, or FFmpeg behavior. Update this skill whenever the workflow contract or common command changes.