Short answer: to save video references for research, capture the file itself while you can still access it, extract stills at the moments you're citing, and record provenance alongside: source URL, retrieval date, and a checksum. Keep an intermediate copy at the delivered rendition rather than a re-compressed one, because analysis needs the pixels the source actually had.
Why a saved copy beats a bookmark
- Link rot is the norm. Hosted video is removed, made private, or re-uploaded with different timestamps more often than it persists.
- Re-encoding destroys evidence. If a claim depends on what is visible in a frame — a gauge reading, a caption, an object's position — a platform's compression pass may have already erased it.
- Timecodes must be reproducible. A citation to 00:14:32 in a file that no longer exists, or whose length changed after a re-upload, is not checkable by anyone including you.
The workflow
Capture the rendition as delivered
Play the item, list what the page fetched, and save the highest rendition without converting it. If the source is segmented, merge with stream copy so nothing is re-encoded — segments and DASH both end this way. The general capture step is in downloading a video from a website.
Extract stills at cited moments
FFmpeg pulls a frame at an exact timecode, so your figure matches your citation:
ffmpeg -ss 00:14:32 -i item.mp4 -frames:v 1 fig-14-32.png
Keep the PNG lossless. Note that -ss placement matters: before -i it seeks by keyframe and is fast; after -i it decodes precisely. For evidence, decode precisely.
Record technical metadata with the file
Run ffprobe -v quiet -print_format json -show_format -show_streams item.mp4 and save the output as a text record next to the video. It captures codec, resolution, frame rate, duration and bitrate — the facts a reviewer might question later.
Fix a checksum and a retrieval date
shasum -a 256 item.mp4 gives you a fingerprint that proves a later copy is the same file you analysed. Write the retrieval date and source URL into a sidecar file; that, not the filename, is provenance.
Store it where the project survives
Project drives get pruned. Keep originals on institutional storage or a second disk, and keep the sidecar files in the same relative layout so a future reader can reconstruct what you had.
Format choices that age well
| Purpose | Choose | Because |
|---|---|---|
| Master copy to analyse | The delivered rendition, remuxed losslessly (MP4 or MKV) | No added generation of loss; MKV tolerates odd track layouts |
| Working copy for editing/annotation | A derived copy, not the master | Keeps the original untouched and verifiable |
| Figures | PNG stills named by timecode | Lossless and traceable to the exact frame |
| Metadata and provenance | Text or JSON sidecars | Readable long after any player is gone |
Container and codec semantics are documented in FFmpeg's formats reference; for browser-delivered material, codec parameters are described in MDN's codecs parameter page.
Working with what platforms permit
Research use and platform terms are not always aligned, and this is where judgement, not technique, decides things:
- Content you may view but that prohibits copying is a grey zone in technique and a bright line in obligations — read the terms, and where they forbid download, request access from the uploader instead. Many will grant it for research.
- Where a source is behind a login you legitimately hold, capture within your own session locally, never by exporting cookies to a third-party converter.
- Human-subject material needs more than a checksum: store it with the consent basis attached, and don't circulate stills of identifiable people beyond what that consent covers.
- Quotation and analysis of a short excerpt is a different act from holding a full copy; if you publish, keep the excerpt minimal and cite the source so readers can check it.
Reproducibility checklist
- Original file with checksum recorded.
- Sidecar with source URL, retrieval date, and ffprobe output.
- Stills named by timecode, each traceable to the master.
- Timecode references in your notes using the master's clock, not a platform player's.
- A second copy in different storage.
That list costs five minutes per source and is what makes a video citation survive review.
FAQ
How do I save research video references locally?
Capture the delivered rendition without re-encoding, extract lossless stills at cited timecodes, and store provenance — URL, retrieval date, checksum — beside the file.
Why keep the original rendition instead of a smaller file?
Because analysis may depend on detail. A second compression pass can remove exactly what a claim about the frame relies on.
Which container should I archive in?
MP4 is the safe general choice; MKV when the stream has unusual track or subtitle layouts. Either way, remux rather than re-encode.
How do I capture a frame at an exact timecode?
With FFmpeg decoding precisely: place -ss after the input so it decodes to that frame instead of seeking to the nearest keyframe.
Can I cite a video I downloaded?
Yes — cite the source URL, the publisher, the publication date, and the timecode, and note that you hold a captured copy with its retrieval date.
Is it acceptable to save video from a login-only research archive?
Capture only what your own authorised access permits, keep it local, and follow the archive's terms; where they forbid copies, request permission rather than working around them.