.png)
Intelli Image Search
Search an image folder using free-text natural language queries powered by Google Gemini AI
Intelli Image Search lets operators search through a folder of images using plain English descriptions instead of face embeddings or keywords. The system uses Google Gemini Vision AI to score every image in the selected folder against the query, returning the top matches with a relevance score and AI-generated reasoning.
Example Queries
- "photo with the highest number of people"
- "image where everyone is smiling"
- "photo taken outdoors in daylight"
- "group photo where someone is wearing a red jacket"
How It Works
- Enter your Gemini API key — the search uses the Gemini Vision API, which requires a valid key from Google AI Studio.
- Select a server folder — choose the image archive to search from the dropdown.
- Enter a query — describe what you are looking for in plain English.
- Start Search — the job runs in the background. Each image is individually analysed by Gemini against the query and assigned a match score from 0 to 10.
- Review results — top matching images are displayed in a gallery with their score and the AI's reasoning for the match.
Scoring
Each image receives a match score (0–10) from the Gemini model based on semantic relevance to the query:
| Score | Meaning |
|---|---|
| 8–10 | Strong match |
| 5–7 | Partial match |
| 0–4 | Weak or no match |
Only the top-ranked images are returned in the results view.
API Key
The Gemini API key is passed client-side at search time and is never stored on the server. It is transmitted securely over HTTPS. Use the eye icon in the key field to toggle visibility.
Limitations
- Intelli Image Search is powered by a third-party API (Gemini) and therefore requires an active internet connection.
- Search accuracy depends on Gemini's vision capabilities — very low-resolution or heavily compressed images may return lower-quality results.
- Large folders increase cost and latency proportionally, as each image consumes one Gemini API call.
- Unlike face-based search, this tool is not guaranteed to be deterministic — the same query on the same images may occasionally return slightly different scores.
Best Practices
- Be specific in queries — "person standing near a window" performs better than "person inside".
- Use image folders with consistent content for best relevance ranking.
- Start with a small folder to verify query phrasing before running across a large archive.