extract-frames-skill

# Extract Frames Sample frames from a video at a constant rate with ffmpeg and save each frame named by its timestamp in seconds (zero-padded so filenames sort correctly). ## Defaults - fps: `3` - format: `jpg` (quality `-q:v 2`) - output dir: `./frames` (created if missing) - filename pattern: `SSSSS.mmm.ext` — seconds, zero-padded to cover the video's duration, with 3 decimal places (milliseconds), e.g. `08177.333.jpg` Ask the user before running if any of these should differ: 1. **fps** — how many frames per second to sample (default 3) 2. **format** — `jpg` or `png` (default jpg; use png for lossless/transparency) 3. **output directory** — default `./frames` 4. Optionally: a specific time range (`-ss` / `-to`) instead of the whole video If the user doesn't specify, use the defaults without prompting further. ## Workflow 1. Confirm the input video path exists (`ls` or `ffprobe`). 2. Get the video duration to determine zero-padding width: ``` ffprobe -v error -select_streams v:0 -show_entries format=duration \ -of default=noprint_wrappers=1:nokey=1 input.mp4 ``` Integer-seconds digit count = `len(str(int(duration)))`, minimum 5 digits. Total printf width = digits + 1 (decimal point) + 3 (decimals). 3. Create the output directory: `mkdir -p frames` 4. Extract frames sequentially with the fps filter (this guarantees frames land at exact `index/fps` second boundaries): ``` ffmpeg -hide_banner -loglevel error -i input.mp4 \ -vf fps=<FPS> -q:v 2 frames/tmp_%06d.<ext> ``` (Drop `-q:v 2` for png.) 5. Rename each `tmp_NNNNNN.ext` to its timestamp. For 0-indexed frame `i`, timestamp = `i / FPS` seconds. Example bash loop (adjust width `W` from step 2, default 9 for `%09.3f`): ```bash cd frames for f in tmp_*.<ext>; do n=$(echo "$f" | sed -E 's/tmp_0*([0-9]+)\.<ext>/\1/') idx=$((10#$n - 1)) ts=$(awk -v i="$idx" -v fps="<FPS>" 'BEGIN{printf "%0<W>.3f", i/fps}') mv "$f" "${ts}.<ext>" done ``` 6. Report frame count and a sample of filenames (first/last few, sorted). ## Notes - Never use ffmpeg's `-frame_pts` for this — it emits pts in the container's internal timebase, not seconds, and isn't human-readable. - The `fps` filter (not `-r` on output alone) is what guarantees evenly spaced, deterministic sample points regardless of the source frame rate. - For very long videos, extraction can take a while; run with a generous or no timeout.