Gemini Backend
Google Gemini Flash API. The default and recommended backend.
Setup
Option 1: API key (simplest)
- Get a free API key at aistudio.google.com/apikey
- Run
eyeroll initand paste the key, or set it directly:
Option 2: Service account (for teams / CI)
For automated or team use, eyeroll supports Google service account credentials:
export GOOGLE_APPLICATION_CREDENTIALS=/path/to/credentials.json
export GOOGLE_CLOUD_PROJECT=your-project-id
export GOOGLE_CLOUD_LOCATION=us-central1 # optional, defaults to us-central1
eyeroll also checks these paths automatically:
~/.eyeroll/credentials.json~/credentials.json./credentials.json
You can also configure this via eyeroll init -- it will ask whether you want to use an API key or a credentials.json file.
Service account auth uses Vertex AI under the hood.
Capabilities
| Feature | Supported |
|---|---|
| Direct video upload | Yes -- File API up to 2GB (API key) or 20MB inline (service account) |
| Frame-by-frame analysis | Yes (fallback for videos exceeding limits) |
| Audio transcription | Yes (native) |
| Text generation | Yes |
| Preflight health check | Yes (verifies model access) |
Model
Default model: gemini-2.0-flash
Override with the --model flag:
Direct video upload
With an API key, eyeroll uses the Gemini File API to upload videos up to 2GB. The model receives the full video in a single request, producing better results than frame-by-frame because it sees motion, transitions, and timing.
With a service account (Vertex AI), videos are sent as inline bytes with a 20MB limit. Videos exceeding this fall back to frame-by-frame automatically.
For videos exceeding either limit, eyeroll falls back to frame-by-frame analysis.
Cost
Gemini Flash is one of the cheapest vision APIs available. A typical 1-minute video costs approximately $0.15, depending on frame count and video length.
Free tier
Gemini API has a generous free tier. For occasional use, you may not need to pay at all.
Install
This installs the google-genai SDK.