Image Search & Semantic Search

Pinpointing targets among vast collections of photos is a core need for professional photographers. Photocatalyst provides three methods — image search, semantic search, and advanced combined filtering — so you can find the exact photo you want among hundreds of thousands in seconds.


Traditional keyword/filename search struggles with massive photo libraries — you can't remember the filename of every photo, let alone manually tag all of them. Photocatalyst automatically extracts visual features and semantic vectors for every photo during indexing, enabling content-based matching without any keywords.


Image search is the most intuitive search method: specify a reference photo and find all visually similar photos.

Suitable for finding photos with the same scene or composition:

  1. Click the "Image Search" icon on the right side of the search bar
  2. Select the "Full Image Similarity" tab
  3. Upload a reference photo
  4. The system automatically analyzes features and returns results sorted by similarity in descending order

Typical Scenarios:

  • Selection decisions: You shot dozens of photos of the same scene and want to see versions from different composition angles
  • Tracing similar photos: After finding a great photo, find all alternative picks in the same group
  • Finding derivative versions: Find multiple versions of a photo before/after processing

Suitable for finding photos with the same local content (e.g., finding where a specific landmark appears across all photos):

  1. Switch to the "Partial Similarity" tab
  2. Add 1~5 reference images
  3. Drag to draw a crop box on each image to outline the region of interest
  4. Confirm, and the system merges features from all regions to perform the search

Typical Scenarios:

  • Find all photos containing the same building/landmark across all photos
  • Find image regions containing specific logos or text
  • Track the use of a specific prop across different shooting sessions

Similarity Threshold Adjustment

The search results page has a threshold slider (0 ~ 1) for real-time adjustment:

Threshold RangeEffectUse Case
0.8 ~ 1.0Only matches highly similar imagesFinding nearly identical bursts/copies
0.6 ~ 0.8Balances precision and recallGeneral search (default 0.6)
0.0 ~ 0.6Includes more images with lower similarityExploratory search, broadening scope

Adjusting the threshold does not re-trigger the search; results refresh in real time.

Right-click on any photo → "Find Similar Images" → "Search Similar Images" (shortcut Ctrl+L) to trigger without opening the search dialog.


2. Semantic Search (Visual Content)

Semantic search lets you describe image content in natural language. The AI understands semantics, not just keyword matching.

How to Use

  1. Select "Visual Content" from the dropdown on the left side of the search bar
  2. Enter subject descriptions, separating multiple subjects with commas
  3. Press Enter to execute

Description Tips

Input ExampleSearch Intent
catPhotos containing a cat
mountain, lake, sunsetLandscape photos with mountains, water, and sunset
wedding, dress, outdoorOutdoor wedding dress photos
building, night, lightsNighttime architecture
food, closeup, restaurantFood close-ups

Practical Advice

  • Use concise and clear subject terms; the more precise the description, the more relevant the results
  • Separate multiple subjects with commas; the system matches each subject independently and ranks comprehensively
  • Supports both Chinese and English descriptions
  • Combine with folder-scoped search to significantly narrow the search scope

Typical Scenarios

  • Client photo selection: A client says "I want the one with the sunset and the beach" — just type it in and find it
  • Portfolio collection: Enter a theme keyword to quickly gather specific types of work
  • Content audit: Check whether there are photos of specific content in the photo library

  1. Click the "Face Search" icon (second from the right) on the search bar
  2. Select a photo containing faces
  3. The system automatically detects faces in the photo and shows thumbnails for selection
  4. Click the target face to jump to the search results page

Use frontal, well-lit, unobstructed face photos for best results. Avoid masks, sunglasses, or extreme profiles.

Search from Preview

In preview mode, right-click a face box on the photo:

  • Find Their Photos: View all photos of that person
  • Find Photos with Similar Faces: Search for similar people based on that face's features

4. Advanced Combined Filtering

Click the advanced search icon on the left side of the search bar to open the precise filtering dialog:

Filter DimensionDescriptionCommon Photographer Scenarios
PathFilter by file pathLimit to a specific project/client folder
Keyword/TagFilter by tagsRetrieve tagged photos
OCR TextFilter by text recognized in photosFind photos containing specific text
CameraFilter by camera modelDistinguish photos from different camera bodies
LocationFilter by GPS locationFind photos from a specific shooting location
File TypeJPEG / RAW / HEIC etc.Find only RAW originals
Rating1~5 stars, supports equal/greater than/less thanFilter high-rated work
Face CountFilter by number of faces in the photoFind single-person/multi-person photos
Time RangeSpecify start and end timeLimit to a specific shooting batch
RecursiveWhether to include subfoldersPrecisely control scope

All conditions can be combined, and results are displayed instantly.


5. Managing Search Results

Search Results Page Features

  • Result count: The page header shows the total number of search results
  • Filename re-filtering: Further filter results by filename
  • Folder scoping: Shows the current search scope; click to clear and return to full-library search
  • Threshold adjustment (image/face search): Real-time slider adjustment
  • Go back: Return to the previous search, supporting step-by-step backtracking

Saving Results

  • Search to album: Save all search results as an album in one click
  • Batch operations: Perform export, tagging, and other batch operations on search results
  • Cross-page selection: Collect photos across different search results

Search Strategy Guide

Finding a specific photo in a massive library:

  Know the filename? → File info search (supports wildcards * and ?)
  Can't remember the filename but have a similar photo? → Image search (full-image similarity)
  Only a vague impression? → Semantic search (describe visual content)
  Remember shooting parameters or time? → Advanced combined filtering
  Photo has a specific person? → Face search

Related chapters: Chapter 4: Search | Chapter 5: AI Features