ClipStays in your storage→SELECTED FRAMES
Frames + your logging fields→THIS MAC
Local reasoning
A model on this Mac processes the selected frames here. A LAN endpoint sends them to another computer.
Keep the big file. Send the useful frames.
A 20 GB camera clip does not need to become a 20 GB upload for AI logging. ClipLogger selects frames that help explain what happens in the clip, then sends those images to the model with the logging fields you want it to fill in. Your original stays where you keep it.
The amount sent depends on the clip, the frames selected and the job. For visual logging, selected images give the model a useful view of the footage while avoiding the full video upload.
Tell the model what you care about
Your logging palette supplies the questions and possible answers. For a game, you might ask for the action, shot size and people involved. For a menu shoot, you might want the dish, pouring or plating action, and shot size. An interview might need the speaker and topic. ClipLogger can also include enabled context such as text seen in the frame, a transcript, and relevant project or roster details.
Rush can receive audio for transcription as well. Selected frames describe what the camera sees; audio helps recover what someone said. Those are different inputs, and the original video is still not the upload.
Choose where the analysis happens
| Route | Where the selected inputs go | What you pay for |
|---|---|---|
| Local on this Mac | A supported model running on your computer | No per-clip service fee; your Mac supplies memory and processing time |
| Your local endpoint | The server address you configure | Your own setup; localhost is this Mac, a LAN address is another machine |
| Your API | Your chosen cloud provider | Eligible ClipLogger paid access plus your provider’s usage bill |
| Rush | ClipLogger’s hosted processing service | Credits for eligible clips; no model-server setup |
Choose the route before starting the batch. Local is useful when you have time to let the Mac work; Rush is useful when you want hosted processing or need your editing machine for something else.
Subject matching happens on-device
ClipLogger groups and compares subject evidence locally, so you can connect appearances to named people, items and places, then review uncertain matches. A person can have face and appearance evidence under the same subject record.
That does not blur people out of an image you choose to send to a cloud model. Selected frames can show faces, and enabled roster or transcript context can contain names. Use a model on this Mac when those analysis inputs need to stay here.
Review the suggestion against the footage
A model sees the selected samples, which can miss a brief gesture or a hidden jersey number. Open the clip, check the proposed values and correct the log. Use a more relevant range when you need the model to focus on one part of a long take.
The result is a log you can use: searchable fields, known subjects and useful ranges. Enable sidecar writing to keep the fuller record with the media, and turn reviewed fields into filenames when you are ready to hand the shoot off.