← All dispatches
Field Reports··5 min read

Local vs. cloud AI color tools

Most AI color tools upload your footage to a server. MOVON runs its AI on a server too, but only color data ever leaves your machine. Where the line sits, and why it matters.

M
Movon Team
MOVON Labs

A lot of AI color tools work the same way under the hood: your footage, or frames of it, gets uploaded to the company's servers, processed there, and the result comes back. It is a reasonable engineering choice. It is also worth knowing about, because "the AI runs on a server" and "your footage goes to a server" are not the same sentence, and the difference is the whole point of this post.

Where "AI color" usually runs

Cloud processing is the default for most AI tools because the models are large and a server with a big GPU is easier to control than every customer's machine. Upload, process, download. For a lot of software, fine.

For footage, it carries quiet costs:

  • Your material leaves your machine. Frames of the project sit, however briefly, on infrastructure you do not control.
  • The upload is the bottleneck. Footage is heavy. The match is gated on your bandwidth, and a long timeline means a long wait before processing even starts.
  • There is a record of your footage. What you uploaded and when, sitting on someone else's storage.

What an NDA makes of that

A lot of professional work is under NDA: an unreleased film, a brand's campaign before launch, a client's confidential event. "We upload frames to our servers for processing" is a sentence that turns a tool into a compliance conversation.

For an editor on that kind of project, "where does my footage go" is the first question about any new tool, and it should be.

How MOVON Match runs

MOVON's AI also runs on a server. That is a deliberate choice: the models that read color and work out the grades are large, and running them server-side is what lets them be as good as they are. A model shrunk down to fit inside a plugin would be a weaker model, and a weaker match.

The line MOVON drew is somewhere else. Not *where the AI runs*, but *what crosses the wire*. When you run a match, the plugin reads the color data from the clips you selected and the reference you chose, and sends only that to the MOVON server. No frames, no footage, no project files. The server compares the numbers, works out one grade per clip, and sends the grades back for the panel to apply.

You sign in with a MOVON account, and the match needs an internet connection. Your footage needs neither, because it never goes anywhere.

Two details worth spelling out, because they sound like exceptions and are not:

Picking a reference file. The panel offers a still or a clip from your drive as a reference, and the button says upload. That is the wrong word. Choosing the file hands its path to the MOVON process running on your own machine, which opens it there, reads its color, and passes on the numbers. The file goes nowhere.

Saving a Look. If you save a match as a reusable Look, one small preview frame is stored with it so you can recognise it later in the list. That frame is the single image that ever leaves the machine, and only when you choose to save a Look. If that is a problem on your project, do not save Looks; the matching itself does not need them.

The boundary was there before the first build shipped. It is how the tool is put together.

What an NDA makes of that boundary

For work under NDA, the question "where does my footage go" has a short answer with MOVON: nowhere. Color data describes how a clip renders color; it is not the picture, and it cannot be turned back into the picture. The unreleased film, the campaign, the client's event: none of it leaves the machine.

The trade-off

Server-side processing costs something, and it is worth being plain about it. MOVON needs a connection and an account. Offline in a basement edit suite, the match waits until you are back online. And a round trip will never be as instant as a local lookup, though at a median of 1.1 seconds per clip across the beta it is close enough that the wait is not what slows the edit down.

What that buys is match quality that does not depend on how new your Mac is, from models too large to ship inside a plugin. For a tool that touches footage under NDA, the deal is: the heavy lifting happens on our GPUs, and your footage never does.

The MOVON Labs team

— End of dispatch N°011
M
Written by
Movon Team
MOVON Labs
The dispatch list

Get the next dispatch in your inbox.

Release notes, color science notes, and the occasional field report. No marketing. Unsubscribe one click.