Fill shot metadata with AI Autofill

AI Autofill looks at one representative frame from each shot and proposes a value for the shot metadata fields you have explicitly switched it on for. It fills those fields and nothing else, and every field arrives with its Autofill switch off.

It earns its keep on details that are visible and classifiable — framing, interior or exterior, roughly what is happening. It is not a substitute for reading your own board. Reviewing what came back is part of the job, not a sign something went wrong.

Every plan, and an internet connection. AI Autofill is included on every plan. The Free plan covers your first 100 shots — a one-time allowance per account, counted by StoryFolder's server, and a shot the model could not tag is not counted — and Pro is unlimited. See Compare plans. Every run needs to reach StoryFolder's servers.

This one is a cloud feature

Your video file is never uploaded. One frame per shot is, along with the field definitions and context described in what gets sent. Transcription runs entirely on your computer; Autofill does not. See Local vs cloud.

Before you begin

• A video that has finished importing and has shots.
• At least one shot metadata field with Autofill switched on. Until then the Autofill action does not appear anywhere in the app.
• A first attempt worth learning from: one field, a handful of shots.

New to custom fields? Start with Custom fields and Field types.

Turn on Autofill for a field

1. Open Preferences (Cmd+P on Mac, Ctrl+P on Windows).
2. Go to Notes & Metadata → Shot Metadata.
3. Find the field you want filled, or create it.
4. Turn on its Autofill switch.
5. Write a one-line field description saying what a good answer looks like.

The description is the cheapest steering you have, and it is sent with every run. Shot size described as "framing of the main subject: wide, medium or close" produces steadier answers than Shot size described as nothing at all.

Switching a field on makes it eligible. It does not fill anything yet, and it does not change values you already have.

Choose your default safeguards

Two rules live in Preferences → Notes & Metadata → AI Autofill and apply to every run.

Autofill on import

Off by default. Turn it on and a video that finishes importing is queued for Autofill across every enabled field, with no dialog and no further prompt.

Leave it off while you are still shaping your fields, working on long videos, or testing a field's description. A forty-minute video is hundreds of model calls from one drag-and-drop.

Only fill empty fields

On by default, and the guard that matters most. With it on, a field that already holds a value is never touched — whatever wrote that value.

"Empty" is per type: an unticked checkbox, an unrated rating, a tag field with no tags, a blank text or number field. A rating of 0 is unrated; a number of 0 is a value.

Overwriting stays possible, but only as a deliberate choice inside a single run — never as a default.

Run Autofill

1. Select the shot or shots you want filled, in the editor or the library.
2. In the inspector, choose Autofill with AI….
3. The Autofill fields with AI dialog opens with every enabled field ticked. Untick the ones this run should leave alone.
4. Choose the scope: This shot, N selected, or All N shots.
5. Choose the write mode: Only fill empty fields or Overwrite existing values.
6. Add Additional context if this run needs steering the field descriptions do not give it — "this is a cooking show; focus on the dish being prepared".
7. Check the count line at the foot of the dialog. It reads N fields · M shots, and it is what the run will actually do.

Overwriting cannot be undone

Choosing Overwrite existing values shows a counted warning — will overwrite N filled values across M shots — before you start it. An Autofill run creates no restore point, so version history cannot bring those values back. Read the count, then start the run.

Then click Autofill N shots.

Run instructions are temporary: they are typed fresh each time the dialog opens and are never saved. Field descriptions persist, which is why anything you will want twice belongs in the description instead.

Follow a run after it starts

The dialog closes as soon as the run is queued. Two things then report it:

• Per-field spinners on the shot cards and in the inspector, on exactly the (shot, field) pairs still waiting or in progress.
• The Autofill card, a small panel that stays with you as you move around the app. It says what the run is doing, shows progress while it works, and offers whatever action makes sense at the time.

In the app's own words:

The card says	What it means
Tagging shot N of M	It is working through your shots. Navigating away is safe
Paused — StoryFolder is busy. Resuming in 4:31.	StoryFolder asked the run to slow down. It resumes itself
Paused — waiting for a connection.	Your computer is offline. Resume now is offered once you are back
Tagged N shots	Finished, with nothing outstanding
Tagged N of M shots. K could not be tagged.	Finished, partly. Retry K shots starts a follow-up run over just those
A failure sentence	The run stopped. The card carries the reason and the action that fits it

A run keeps going when you navigate away, and it survives quitting and reopening StoryFolder. Closing the window does not cancel it.

Anything that does not resolve itself is covered in Troubleshoot AI Autofill.

Review what came back

Review a small sample first

Take the beginning, the middle and the end of the board rather than the first screenful — the first ten shots of anything are the easiest ten shots in it.

Then go looking for trouble on purpose: transitions, graphics and titles, very dark shots, a subject that is partly hidden, and frames where two subjects are equally plausible. If a field survives those, it will survive the rest.

Review one field at a time

For a field with a fixed set of answers, filter the board to one value at a time — Shot size is Wide, then the next option — and read the queue. Wrong values stand out against their neighbours in a way they never do in shot order.

Look for gaps as well as errors: an option that came back for nothing at all usually means the description is not describing it.

Correct values in place as you go. When the field behaves, widen the run.

Open Text and Tag fields have no queue to filter into, so take a fixed sample instead and read it properly.

What each field type allows

Field type	What comes back	Held to your options
Dropdown	One of the options you defined	Yes
Rating	A whole number, 1–5, or 0 for unrated	Yes
Checkbox	Ticked or unticked	Yes
Number	Any finite number	Yes
Tag	Short tags, including ones you never listed	No — open by design
Text	Anything	No — a pass-through

That split is the one to hold on to. A constrained field can only come back with an answer from your own list; an open one can come back with anything. Either way, a value that fits the field is not evidence that it is the right value — "Wide" is a legal answer to every shot on your board.

Full definitions live in Field types.

What gets sent

For each shot in the run:

• One representative frame — the same frame the board shows as that shot's poster.
• The fields you ticked: their names, their types, the options of a dropdown or tag field, and the descriptions you wrote.
• This run's additional context, if you typed any.
• What the project already knows: its title and description, and the video metadata you have filled in — so "which product is this?" has an answer before anyone looks at a frame.

Your source video is never uploaded. That is a narrower promise than "nothing leaves the computer", and the difference is the point: frames and field context do leave, to StoryFolder and on to the AI model. Treat a confidential shot note the way you would treat anything you paste into a chat window.

See Local vs cloud and Privacy & sharing.

What AI Autofill is good at

Things a single frame shows. Depending on how you have described your fields, that usually includes:

• framing and shot size
• interior or exterior, day or night
• roughly how many people are in shot
• broad action or subject
• obvious technical problems — out of focus, badly exposed

Treat that as a starting list, not a supported taxonomy. Your fields are yours; what Autofill can do with them depends on what you asked for.

What to enter yourself

Anything that is true of the job rather than visible in the picture: client, job number, rights and release status, approval state, who booked the day, where the footage came from.

• Facts about the whole project belong in video metadata, typed once.
• A known value that many shots share is a copy and paste job, not a model's guess.

Limits worth knowing

• Results can be wrong, and results can be blank. Both are normal outcomes.
• One frame is not the whole shot. Action that starts after the poster frame is action Autofill never saw.
• A rejected value is silent. An answer outside a field's allowed set is dropped and nothing is written, so an empty cell does not tell you whether the model declined or was overruled.
• Overwrite replaces human work. With Overwrite existing values chosen, a value you typed is a value it can replace.
• There is no undo. Autofill writes directly and records no restore point. Version history has nothing to roll back, and Cmd+Z will not reverse a run. Your guards are Only fill empty fields and running small before running wide.
• Nothing marks a value as AI-written. In the cell it looks exactly like one you typed.
• It does not know who anyone is. No identity, no consent, no rights status, no facts from outside the frame and the context you sent.
• The Free plan covers 100 shots. On the Free plan the Autofill fields with AI dialog shows how many of your 100 shots are left. A run larger than what is left fills the first shots and then stops with an upgrade prompt; the shots it filled stay filled. Pro is unlimited — see Compare plans.

Related pages

• Troubleshoot AI Autofill
• Custom fields · Field types
• Video metadata
• Copy and paste metadata
• Version history
• Local vs cloud

Next step

→ Troubleshoot AI Autofill