Semantic search: find images by content with AI
Discover the power of semantic search in Polimake. Find images by typing natural descriptions, without needing to remember file names.
On this page
Describe the visual content you need, even if you do not remember the file name. Semantic search compares your description with images in the library.
What is semantic search?
Semantic search uses artificial intelligence to:
- Understand the visual content of your images
- Search by concept instead of just by text
- Find related images even if they don't match exactly
- Understand natural descriptions that you write in everyday language
It's like having an assistant that "sees" your images and finds them when you describe what you're looking for.
How it works
The process behind the scenes
When you upload images to Polimake:
- Each image is analyzed automatically with AI
- Embeddings are generated (vector representations) of the content
- They're stored for fast searching later
When you search:
- Your query is converted into a similar embedding
- It's compared with the embeddings of your images
- The results are ranked by relevance
- The most relevant images are shown first
Automatic analysis
Every image you upload is analyzed to extract:
- Objects and elements visible in the image
- Concepts and themes represented
- Style and aesthetic (colors, composition, mood)
- Context and potential use
This information is used to make search smart and contextual.
How to use semantic search
Accessing semantic mode
To use semantic search:
- Click the search bar in the header
- Select "Semantic" mode (brain or AI icon)
- Or just start typing, semantic mode is the default
Visual indicator: The icon in the search bar shows you which mode is active.
Writing effective searches
For best results, write the way you'd speak:
Examples of effective searches:
- ✅ "beach images at sunset"
- ✅ "photos of Italian food"
- ✅ "people smiling in an office"
- ✅ "fashion products on a white background"
- ✅ "nature with mountains and lakes"
Avoid overly specific searches:
- ❌ "1920x1080 jpg image with a beach"
- ✅ "beach"
Tip: Be descriptive but natural. Write as if you were describing the image to another person.
Waiting for results
After typing your search:
- Press Enter or wait a moment
- The search is processed (it may take a few seconds)
- The results appear ranked by relevance
Processing indicator: You'll see an indicator while the search is being processed, especially for complex searches.
Practical examples
Illustrative queries and possible matches. Results depend on the available material; these are not observed customer searches or retrieval guarantees.
Case 1: Search by concept
Search: "summery images with happy people"
Possible matches:
- Beach photos with people
- Images of a picnic in a park
- Photos of outdoor festivals
- Other images that may suggest a summery mood
Doesn't require: That the file be named "summer" or "happy"
Case 2: Search by style
Search: "professional corporate photography"
Possible matches:
- Office images
- Professional portraits
- Photos of business meetings
- Images with a corporate style
Doesn't require: Specific tags or descriptive names
Case 3: Search by specific content
Search: "Italian food on a wooden table"
Possible matches:
- Pasta dishes on rustic tables
- Pizza in an Italian setting
- Italian food in the right context
Understands: The full concept, not just individual words
Advantages over basic search
No need to remember names
With semantic search:
- You don't need to know what the file is called
- Describe what you're looking for and find relevant images
- It works even if the names are generic (IMG_1234.jpg)
Finds related concepts
Semantic search:
- Finds variations of the same concept
- Relates similar concepts (beach → ocean → sand)
- Understands synonyms and related concepts
A more natural search
You can search:
- The way you'd speak: "nature photos"
- With context: "product images for e-commerce"
- By style: "minimalist photography"
Best practices
Be descriptive but concise
For best results:
- Include key elements: What's in the image
- Add context: Where, when, style
- Avoid being too specific: Let the AI find variations
Examples:
- ✅ "beach sunset people"
- ✅ "Italian food restaurant"
- ❌ "1920x1080 jpg image beach sunset people smiling blue shirt"
Combine concepts
You can combine multiple concepts:
- "fashion product white background": Finds fashion products with a white background
- "nature mountains lake": Finds landscapes with these elements
- "people office meeting": Finds corporate scenes
Use natural language
Write the way you'd speak:
- ✅ "food photos"
- ✅ "nature images"
- ✅ "products on a white background"
- ❌ "food images jpg"
- ❌ "nature photos png"
Comparison with other modes
Semantic vs. Basic
Semantic search:
- Searches by visual content
- Doesn't require knowing the name
- Finds related concepts
- Takes a few seconds to process
Basic search:
- Searches by file name
- Requires knowing the name
- Exact text search
- Text matches in the name
When to use each: Use semantic when searching by content, basic when you remember the name.
Semantic vs. Reasoned
Semantic search:
- Search by visual/conceptual similarity
- Fast and efficient
- Good for discovery
Reasoned search:
- Uses advanced AI reasoning
- May help compare another selection for complex queries
- May take longer
When to use each: Use semantic for most searches, reasoned for very specific or complex cases.
Semantic search results
Ranking by relevance
The results are shown:
- Ordered by relevance: The most relevant first
- With a similarity score: Indicates how well they match
- Grouped visually: Easy to scan
Interpreting results
The results show:
- Images relevant to the concept you searched for
- Variations of the same concept
- Related images that may be useful
Tip: Review the first results, they're usually the most relevant. If you don't find what you're looking for, refine your search.
Limitations and considerations
Requires prior analysis
For semantic search to work:
- The images must be analyzed (this is done automatically on upload)
- It may take time if you've just uploaded many images
- Wait for the analysis to finish before searching
Very specific searches
For extremely specific searches:
- It may not find exact matches
- Consider using reasoned search for complex cases
- Or combine it with basic search if you remember part of the name
Frequently asked questions
Can I find an image without knowing the file name?
Yes. Semantic search compares your description with the visual content of the images, so it works even if the file has a generic name like IMG_1234.jpg. Type what's in the picture, for example “Italian food on a wooden table.”
When can I search for images I just uploaded?
Once their analysis finishes. Polimake analyzes each image automatically on upload, but it can take a few minutes, and longer if you've uploaded many at once.
What's the difference between semantic search and basic search?
Semantic search looks for visual content and concepts, even if you don't know the file name, and takes a few seconds to process. Basic search looks for text matches in the file name. Use semantic search to describe what you're looking for and basic search when you remember the name.
How do I write a query to get better results?
Describe the image the way you'd describe it to another person: the key elements plus some context or style, like “fashion products on a white background.” Avoid technical details like format or resolution, and if what you want doesn't show up, refine your terms based on the results.
See also: reasoned search.
Next steps
- Try semantic searches with natural descriptions
- Experiment with different concepts to see what it finds
- Combine concepts for more specific searches
- Compare results with basic search
- Integrate it into your regular workflow