There is no single rate, because the bill is two numbers you control multiplied by one you don't: how many hours go in, how many times each hour gets processed, and what the vendor charges per unit. Anyone who quotes a per-hour figure without asking how often you re-index has quoted you half the cost. The useful way to approach this is to price the parts separately, then decide which parts you actually need.
What "indexing" is paying for
The word covers several different pieces of work, and products bundle different subsets of it. There is a pass over the audio to get the spoken words. There is a pass over the picture to get what is visible. There is the embedding and storage of whatever those passes produced, which is the part that keeps costing money after the processing is done. And there is the compute at question time, when someone actually searches.
A product that only reads speech is doing less work than one that also reads the frame, and it will be priced accordingly. This matters before you compare numbers, because "indexed" on one invoice and "indexed" on another can describe different things. The breakdown of the audio side sits in the thousand-hour transcription estimate, and the embedding side, which is where the surprises usually are, is worked through in the cheapest route to embeddings and timestamps.
The hours you do not need to index
Most archives contain a lot of footage nobody will ever look for. Reels of setup, duplicate uploads, and the long tail of material that is out of date the year it was shot. Processing cost scales with hours, so the first lever is not the rate. It is deciding what goes in. A rough rule that survives contact with reality: index what someone has asked for at least once, plus whatever is still on brand and still true.
Re-indexing is the line nobody budgets for
The one-time cost becomes periodic more often than people plan for. A better model arrives, the team starts asking about things on screen when the original pass only covered speech, or you move to another product and pay for the whole archive again. Ask any vendor what happens to already-processed footage when their model changes, and who pays for it.
This is why the shape of the processing model matters as much as the rate. On Vivu, each video is processed when it goes into a project, so adding footage and running searches show up as separate lines rather than one that compounds every time somebody asks a question, and the tier limits themselves are listed on its pricing page.
Question time is its own line
Search is not always a lookup. In our own runs, a precise pass ran as a job you wait for rather than something instant, and the waiting is a hint about what is happening underneath: real work per question, which usually means real cost per question. A broad sweep over the same library is cheaper and returns less, often just which files might be relevant.
Whether that lands on your invoice depends on the product, but it should land in your estimate either way. A team of four people each asking a handful of questions a day is a different customer than a team that searches twice a month, even with the same archive. Automatic tagging products price the same tension from a different side, which the tagging comparison gets into.
When indexing isn't worth paying for
If the library is small, or new enough that the people who shot it still work there, the retrieval problem is usually a naming problem and indexing is an expensive way to skip a conversation. The same is true for a one-off: paying to process an archive so you can find one clip is worse value than scrubbing for it. Indexing earns its cost when questions repeat, come from people who were not on the shoot, and are about content rather than filenames.
How to arrive at a number you can defend
Write down four figures before you talk to anyone: hours you would ingest now, hours added per month, searches per person per week, and how many people need access. Put a re-index of the whole archive in the plan once, whether or not you expect it. Most of the disagreement about whether video indexing is expensive comes from comparing a rate against an archive nobody has sized yet, and that is a question you can answer in an afternoon without a vendor in the room.
FAQ
Is video indexing a one-time cost or a subscription?
Both, usually. The processing of a given file happens once and is priced as work, while storage of the footage and its index, along with access for your team, continues as long as you keep the library. The part that turns a one-time cost into a recurring one is re-processing, which happens when models change or when you start asking a kind of question the first pass did not cover.
Why does indexing some footage cost more than indexing other footage?
Length is the dominant factor, since the work is proportional to hours of material. Beyond that, what gets run over each hour changes the total: a pass for spoken words alone is less work than one that also covers what is on screen. Footage with no usable audio gets no benefit from the speech side, so paying for it is worth checking before you upload a silent archive.
Do we need to index everything we have ever shot?
No, and trying to is the most common way this gets expensive. Start with the material someone has already asked for twice, and with anything still accurate enough to reuse. Older footage can go in later if the searches keep pointing at it, which is cheaper than processing it all on the chance that it matters.
How do I estimate this before contacting a vendor?
Measure your archive in hours rather than terabytes, since processing is priced by time and not by file size. Add your monthly growth, an honest guess at searches per person per week, and the number of people who need access. Those four numbers turn any quote into a yearly figure, and they also show you quickly whether the problem is worth buying a product for.