Local vs Gemini: choosing a processing mode

FrameSeek can index footage using local AI models or the Gemini API. Here is how they compare and when to use each.

FrameSeek supports two modes for generating visual descriptions of your footage: Local (default) and Gemini API. You can switch between them at any time in Settings. This guide covers the differences and when each makes sense.

Local mode (default)

Local mode runs AI directly on your Mac. No internet connection is needed and no data leaves your machine.

On macOS 27 and later, local mode uses Apple Intelligence which is the on-device model built into macOS. This means:

  • No model download, it ships with macOS, so you can start indexing immediately
  • Low memory usage, runs on the Neural Engine, not your GPU
  • ~2.5x faster than the previous local model

On macOS 15–26, local mode uses an open-source vision model (Qwen) that runs on your GPU. This requires a one-time ~1.5 GB model download and uses ~5 GB of RAM during indexing.

Regardless of macOS version:

  • Speed: Slower than Gemini cloud, but fully private. On macOS 27, roughly 1 hour of footage takes about 20 minutes. On older macOS, about 50 minutes.
  • Cost: Free.
  • Privacy: Nothing is sent to the internet. All processing happens on-device.
  • Quality: On par with Gemini. The local models produce descriptions that work just as well for search.

Local mode is the right choice when privacy is a priority, when you have a manageable library, or when you don't want to set up API keys.

Gemini API mode

Gemini mode sends sampled images from your videos to Google's Gemini API for visual descriptions. This only happens once, during indexing. After that, all search is local and no data is sent anywhere. Transcription and search also always run locally regardless of mode.

  • Speed: Much faster than local, especially on large libraries. Processing is offloaded to Google's servers.
  • Cost: Paid, based on the amount of footage and the model you choose. The default model costs roughly $0.04 per hour of video. You can pick a different model in Settings. FrameSeek shows the estimated cost before you start indexing, so there are no surprises.
  • Privacy: Your full video files are never sent to Google. Only sampled images are shared for generating descriptions. Audio, transcripts, and search queries never leave your Mac. Requests go directly from your Mac to Google's Gemini service. FrameSeek does not proxy, process, or have any access to your API calls.
  • Quality: On par with local. Both modes produce descriptions that work equally well for search. We benchmarked multiple Gemini models and found no meaningful quality difference for frame description.

Gemini mode is useful when you have a large library and want to get through indexing quickly without tying up your Mac.

How to switch

  1. Open FrameSeek and go to Settings (Cmd+,).
  2. Under Describe Mode, choose Local or Gemini.
  3. If you choose Gemini, enter your API key. You can get an API key from Google AI Studio. We recommend a paid key since free tier rate limits are too low for indexing video libraries. You pay Google directly for any usage. FrameSeek does not charge for or profit from Gemini usage.

You can switch between local and Gemini for each indexing run. If you've added a Gemini API key in Settings, you'll see the option to switch modes when starting a new indexing run. Already-indexed footage keeps its existing descriptions regardless of which mode was used.