Your own footageGuide
How to organize video footage for ads: sort every clip by the job it does
Short answer
Sort footage by the job a clip does in an ad, such as hook, demo, before and after, reaction or call to action, not by shoot date or folder. Map every file with a plain listing first, give clips names that describe them, and keep talking heads apart from silent b-roll.
A folder named for a shoot date doesn’t help when the next ad needs a hook or a before-and-after shot. This guide is how we organize video footage so that a new ad can be built from it, and it fits a small brand’s drive as well as ours. It is the same logic as the library on our page for brands with their own footage. For the whole approach of starting from the footage you already have, see An AI ad workflow for small brands.
Map before you sort
Start with one recursive listing of everything: file names, sizes and a hash for each file. It costs almost nothing, uses no AI, and finds exact duplicates on its own. On Google Drive, the files resource in the Drive API gives each file an ID, a name, a MIME type and a size, plus an MD5 checksum for files with binary content, which covers video files.
In a pilot we ran in October 2026 for a brand outside our own, an AI connection that crawled the folders one by one was slow and only reached the first few levels. A file tool listed the whole drive in one pass. Three rules came out of it:
- Classify by file, not by folder name. Use the extension, the size and the camera name. A folder name says what someone meant to put there, not what is in it.
- Key everything on file ID and hash. One path can hold several different files, and byte-identical copies sit under many paths.
- Don’t read the modified time as the shoot date. Drive documents it as the last time anyone modified the file, which is not when it was filmed.
In Bring your own content Early access, we run this mapping with you, read-only. Connect buttons Coming would let you do it without us. Early access: we set it up with you. Coming: we’re building it.
Sort by job, not by shoot
A clip’s bucket is its function in an ad, not its subject. Footage from one shoot day lands in several buckets, and that is the point.
| Bucket | What the clip does in an ad |
|---|---|
| Hook | Opens the ad and stops the scroll |
| Explanation | Someone explains something, usually a talking head |
| Demo | Shows the product in use or how it works |
| Before and after | Shows the change |
| Reaction | A face reacting, without speech |
| Call to action | Tells the viewer what to do next |
| End card | Closes the ad with the brand |
Source: our own production rules, October 2026.
Look, product, text on screen and speech are separate fields, not buckets. In that pilot, the labelers agreed on function far more often than on look or product, so we have the owner confirm the look in review. We use two looks for real footage: phone footage (UGC) and a professional shoot, where you can tell it wasn’t filmed on a phone. Generated clips get a third.
One shoot day
- HookA reveal in the first second
- DemoThe product in use
- Before and afterThe change
- Call to actionThe real form
Keep the ad function separate from appearance, visible action, framing, duration and source. A short name identifies the clip; structured fields make it searchable without rewriting that identity when a label changes or the file moves. Stable video asset IDs and readable names separates asset identity from descriptive fields, so this organizing guide can stay focused on the folder and cutting decisions.
Cutting rules
When footage is cut into clips, these rules apply:
- Cut at scene changes, never inside a transition or a flash.
- Cut talking heads by sentence, using word times, and never in the middle of a word.
- Remove black frames and fades.
- Keep clips at least one second long, and b-roll at about five seconds at most.
- A clip has to make sense: a clear purpose, and the product in view within the first one to two seconds. Footage of someone just walking around is dropped.
Talking heads get their own folder. A clip where someone speaks on camera is not b-roll, so it never sits among the silent clips. A simple structure is enough: look/talking_heads/ for clips that speak, and look/broll/bucket/ for silent ones, with the look and the bucket from the fields above.
Cutting a talking head by sentence
- 1Cut between words
- 2Cut between words
Flat log footage and unstabilized handheld clips need cleaning before they are cut, which How to repurpose video ads covers.
Use rules for file facts, then let a model suggest only the labels that require interpretation, with a person confirming uncertain or consequential choices. In our October 2026 pilot, a camera field was empty for most clips; making it mandatory would have created invented data. AI tagging for a video clip library covers the closed vocabulary, evidence frames and review route.
Rate with three choices
Every clip gets one of three answers: keep, favorite or archive, plus an optional comment such as shorter or wrong bucket. We chose three choices over five stars on purpose. Stars on hundreds of clips cause decision fatigue, and the difference between three and four stars never gets used for anything. It should work from the keyboard, and delete should mean archive. The keyboard version is what we’re building as Rating inside Turntake Coming.
Keep it growing
Log which clips each ad used, so the library records its own reuse and no clip is leaned on too hard. A b-roll library for ads covers rotation. Your old ads are another source of new clips, and How to repurpose video ads covers how to find the clean footage behind them. For the part of the library that is missing, see AI b-roll for product ads.
Who wrote this We made the ads first, for SmartSolar in Denmark, Sweden and Ireland. Then we built Turntake. Where Turntake comes from
FAQ
What’s a good way to organize video files for editing?
Sort clips by the job they do in an ad, such as hook, demo, before and after, reaction or call to action, not by shoot date. Start with a plain listing of every file with name, size and hash, give each clip a name that describes it, and keep talking heads in their own folder, apart from b-roll.
How should I name video clips?
Name each clip for what it shows, not for the shoot: who or what is in the scene, then the action. The file name is the clip’s ID, so keep it unique. Put the look, the bucket and the other details in separate fields instead of cramming them into the name.
Should talking heads and b-roll share a folder?
No. A clip where someone speaks on camera is not b-roll. We keep talking heads in their own folder and use them only in ads with the creator’s own audio. Voiceover edits use only b-roll that does not talk, so mixing the two makes it easy to put a speaking mouth under a different voice.
Can AI tag my video clips?
Yes, as a first pass. Use rules first, a model where the rules stop, and a person last. The model works from a few frames and a line of transcript per clip, not from the video, so treat its tags as suggestions and confirm the ones that matter. Never make a field mandatory that nothing can fill.
How long should a b-roll clip be?
In our house rules for short vertical ads, at least one second and about five seconds at most. Those are our own limits, not a standard. A clip should also make sense on its own: a clear purpose and the product in view within the first one to two seconds, or it doesn’t belong in the library.
Sources
- Google Drive API: the files resource checked 10 October 2026
Facts from the sources below were checked on the date shown. Pages change, so read the source again before you rely on it.