Why Local Video AI Hits a Ceiling
Desktop tools are bounded by the machine they run on. Here's where that ceiling sits - and when a shoot is big enough to hit it.
Desktop video AI works. You install it, point it at a folder, and it finds faces in your footage without anything leaving your hard drive. For a personal library on one computer, that is a good trade.
But every desktop tool inherits the limits of the machine it runs on. That is not a flaw in any particular app - it is the architecture. The question is whether your shoots are big enough to reach it.
What local processing gets right
Nothing leaves your drive. There is no upload wait, no subscription, and no third party holding your footage. If your work is confidential, or your internet is slow, that matters more than any feature comparison.
For a stable library on a single machine, it is genuinely the right answer.
Where the ceiling is
Your computer does the work, and it can only do one thing at a time. Face recognition is compute-intensive. The GPU chewing through your footage is the same GPU you need for the edit. On a real deadline you are choosing between indexing and cutting, and the indexing pass is measured in hours.
The footage has to be on that machine. If a shoot lives on a NAS, a second workstation, or a client's drive, you copy it across first. For an eight-hour multi-camera shoot that copy is its own afternoon.
It searches, it does not cut. A local tool tells you where a person appears. Turning that into deliverable clips is still a manual export pass. If your next step after finding a clip is exporting it by hand, the search only solved half the problem.
One machine means one person. A second editor, or a client who wants to browse their own footage, has nowhere to go. Local tools have no answer to this because there is no server to ask.
The hardware sets the ceiling, not the software. A faster card raises it. It does not remove it. Every year your cameras record more data than the year before, and the machine under your desk does not.
What changes when the processing moves off your machine
Tabor8 runs the recognition pass on a cloud GPU. You upload from any browser on any device, the job runs on hardware that is not your edit machine, and results come back as trimmed, named clip folders ready to import into your NLE.
The practical difference is that the indexing pass stops competing with your edit for the same hardware, and stops being bounded by it.
| Local processing | Tabor8 | |
|---|---|---|
| Runs on | Your machine | Cloud GPU (A10G) |
| Access | The computer it is installed on | Any browser, any device |
| While it runs | Your edit machine is busy | Your edit machine is free |
| Output | A search index | Actual clip folders |
| Team access | No | Yes |
| Camera formats | Whatever the OS decodes | Any format |
| Privacy | Stays on device | Embeddings deleted post-scan |
| Cost model | Typically a one-time licence | Free tier plus subscription |
When local is still the right call
If you work from one machine, your footage never leaves the studio, and you want a permanent licence with no recurring cost, a desktop tool fits. The ceiling only matters once you are pressed against it.
When the cloud wins
If you shoot more than one machine can index overnight, work across locations or with a second editor, need your edit workstation free while footage is being sorted, or want clips delivered rather than merely found - the processing needs to happen somewhere other than your desk.
That is the whole argument. Not that desktop software is bad, but that a shoot eventually produces more footage than one computer was designed to handle, and at that point where the work runs matters more than which app runs it.
Try the cloud approach.
No install. No hardware requirement. Upload a shoot and come back to sorted folders.
Apply for beta access →Tabor8 is in private beta, New Zealand only.