Stop paying to search your own footage

A footage platform can bill you for two different things: the work, and the doorway. The work is compute, GPU minutes spent transcribing interviews, describing shots, grouping the same face across a season of footage. The doorway is access, the monthly fee that keeps you permitted to read what the compute produced. Pay for the work whenever it saves you time; it is honest labor and it is priced like labor. Never pay for the doorway, because behind the doorway is material you already made.

I watched a team learn the difference the expensive way: three seasons tagged into a cloud library, every interview transcribed, every athlete tagged, every select starred. Then budgets moved and the subscription lapsed. Forty terabytes came home as filenames. The files survived; the understanding stayed behind, in a database they can no longer query.

Where the analysis lands

Read the pitch of most footage-AI platforms and the shape repeats: upload your media, we analyze it, now it is searchable. The word doing the quiet work is where. Searchable in their app, through their search box, out of their database, in their schema, for as long as your card is current. The analysis itself was real work, worth paying for once. But by month seven the recurring bill is not buying analysis; it is buying continued permission to read output that already exists, derived from footage you shot, corrected by people you paid. Meanwhile the archive is the part of your business that appreciates. A drive of 2019 footage is worth something in a sequel year, a retrospective, a b-roll request, but only if it can still answer questions. When the vendor holds the answers, the useful life of your archive is capped at the life of your subscription, which is a strange thing to agree to for the one asset you planned to keep forever.

Compute is a cost, access is rent

I want to be precise here, because this is not an anti-cloud argument. ClipLogger sells cloud compute: Rush runs your batch on parallel workers with larger models than a laptop holds, and it is faster and reads hard frames better. Renting a datacenter for an afternoon is a fair trade, and I take the money gladly. The line is what happens to the results. Rush results land in the same open sidecars as a local run, beside your files, marked with which engine produced them, and they stay there when the account closes. The pricing is flat because the work is the product: one credit analyzes one clip, up to fifteen minutes, cost shown before every run, and the credit packs never expire, because unused compute is not a lease that should rot. You pay to go faster. You never pay to get your own work back, because it never left your disk in the first place. That is the whole distinction: compute is a cost you incur once, access is rent you pay forever.

The exit test

Run this on any tool you are evaluating, mine included, before the first card goes in. Imagine cancelling today, then name, specifically, what you hold tomorrow. The media files, yes, presumably. Now the derived layer, which is where the years of value live: transcripts, tags, the people groupings, ratings, selects, notes. Ask the vendor four things and accept only boring answers: where the derived data lives, beside your files or in their database; what format it is in, documented or proprietary; what a ten-line script can read after their product is gone; and what exactly stops working the day you stop paying. A trustworthy answer to the last one names capabilities: speed, sync, collaboration, support, storage. An answer that includes your own metadata is a ransom note with a pricing page. Vendors do not volunteer this, and the demo will never show you the cancellation, so ask in writing and keep the reply.

What the boring answer looks like

On disk, the right answer is almost disappointing: a plain JSON sidecar beside every clip carrying its transcript, tags, confirmed decisions, and the provenance of each value; standard exports, XMP, Resolve CSV, FCPXML, ALE, for whatever your pipeline already reads; and a local app that is free forever with no account, so search never had a meter on it to begin with. Delete ClipLogger and the folder still explains itself to a text editor. The guarantee is structural, not contractual: metadata that lives beside the files in a documented format cannot be held hostage by anyone, including the people who wrote the app. Subscriptions should buy you speed while they run and cost you speed when they end, nothing else. Anything that costs you your own decisions on the way out was never a subscription; it was a lease on your memory.

If cancelling would make your footage unsearchable, you were never searching your footage, you were searching their database.