# Changelog of Video Frame Extractor & Scene Chunks for RAG (`mmdincer/video-frame-extractor`) Actor

- **URL**: https://apify.com/mmdincer/video-frame-extractor/changelog.md
- **Full Actor documentation**: https://apify.com/mmdincer/video-frame-extractor.md

## Changelog

### 1.0.9 — 2026-11-07

- A video longer than `maxDurationMinutes` is no longer downloaded in full before analysis. The Actor copies the leading minutes (ffmpeg `-t` before `-i`) and stops. The prefilled Sintel file is 77 MB and about 15 minutes; downloading all of it used the whole 5-minute run, so no scene was written. The file duration is still read from the remote header and the row stays `truncated` when only the leading part was analysed.

### 1.0.8 — 2026-11-04

- Documentation only: shorter answer at the top of the README, key terms for scene detection and video to RAG, and cross-links to the PDF and audio transcription Actors as one RAG family. No code or pricing change.

### 1.0.0 — 2026-11-03

First release.

- One chunk row per scene: keyframe JPEG, start and end time (seconds and `HH:MM:SS`), transcript text and token count (cl100k_base).
- One video row per video: length, resolution, codec, scene and chunk counts, transcript source, billed units, contact sheet link. Failed videos write a row with a reason and are not charged.
- Scene detection with ffmpeg (5 fps, 320 px), minimum and maximum scene length, scene cap, and merging of neighbouring scenes that show the same shot (difference hash plus average colour). Fixed-interval mode for footage without cuts.
- Transcript from a subtitle file (`.srt`, `.vtt`), from a dataset of another Actor (`transcriptDatasetId`, matched on `url`, aligned through `segments`) or from a text subtitle track inside the video. Nothing is transcribed.
- Reads video links from another Actor's dataset (`datasetId`).
- Direct video file links only; video platform pages are refused with an explanation. A 403 or 429 is reported as `blocked` and never retried or worked around.
- Pay per event: scenes returned and minutes analysed. Keyframes, contact sheets and extra rows for long speech are free.