Skip to content
← Documentation

Advanced AI search: reasoned search

Master reasoned search in Polimake. Use advanced AI reasoning to find images that match complex, specific concepts.

Available inCloud
Polimake~6 min read
On this page

Reasoned search combines semantic search with AI reasoning for queries with several conditions. Learn when to use it and how to phrase a query.

Reasoned search is a two-step process:

  1. Initial semantic search: Finds relevant candidates using embeddings
  2. AI reasoning: An advanced AI model analyzes and re-ranks the results based on deep contextual understanding

The result is a search that not only finds similar images, but also understands the context and intent behind your search.

How it works

Two-stage process

Stage 1, Semantic search:

  • Your query is converted into an embedding
  • Candidate images are found by similarity
  • An initial set of results is obtained

Stage 2, AI reasoning:

  • An AI model analyzes the candidates
  • It compares the content of each image with your search
  • It re-ranks the results by contextual relevance
  • It reorders candidates according to the query

The advantage of reasoning

AI reasoning makes it possible to:

  • Understand context: Grasp the intent behind your search
  • Relate concepts: Connect related ideas intelligently
  • Prioritize relevance: Show what you really need first
  • Filter out noise: Eliminate results that only match superficially

Ideal cases

Reasoned search is best for:

  • Complex searches with multiple related concepts
  • Abstract concepts that require interpretation
  • Very specific searches where precision is critical
  • Cases where simple semantics isn't enough

Usage examples

Complex search:

  • "images that convey professionalism and trust for a corporate presentation"
  • Reasoned search understands the combination of concepts and the context of use

Abstract concept:

  • "photographs that capture the essence of Mediterranean summer"
  • It understands not only visible objects, but also the "feeling" or "essence"

Specific search with context:

  • "luxury products on an elegant background for a premium campaign"
  • It combines multiple criteria (product, style, background, purpose)

Activating the mode

To use reasoned search:

  1. Click the search bar
  2. Select "Reasoned" mode (advanced brain or AI+ icon)
  3. Type your search as you normally would
  4. Press Enter and wait for the results

Visual indicator: The icon in the search bar shows that you're in reasoned mode.

Writing effective searches

To take advantage of reasoning:

Be descriptive and contextual:

  • ✅ "images of tech products on a minimalist background for a website"
  • ✅ "nature photographs that convey calm and peace"
  • ✅ "professional corporate images for a company presentation"

Include the context of use:

  • Mention what you'll use the image for
  • Describe the desired mood or feeling
  • Specify the required style or aesthetic

Combine related concepts:

  • Don't be afraid to be specific
  • Combine multiple criteria
  • Describe the intent, not just objects

Waiting for results

Reasoned search:

  • May require more processing; timing depends on the query and service
  • Shows progress while processing
  • Returns candidates to check against your criteria

Compare the results: additional processing does not guarantee a better selection for every query.

Comparison with other modes

Reasoned vs. Semantic

Semantic search:

  • Fast (seconds)
  • Good for discovery
  • Based on visual/conceptual similarity
  • Ideal for most searches

Reasoned search:

  • May require more processing
  • Reorders candidates according to the query context
  • Uses advanced reasoning
  • Ideal for complex searches

When to use each: Use semantic for quick, general searches, reasoned for complex searches where precision matters.

Reasoned vs. Basic

Basic search:

  • Focused on text matches
  • Searches by file name
  • Requires knowing the name

Reasoned search:

  • Focused on content and context criteria
  • Searches by content and context
  • Doesn't require knowing the name

They're complementary: Use basic when you remember the name, reasoned when you're searching by a complex concept.

Reasoned search results

Improved ranking

The results are shown:

  • Re-ranked by contextual relevance: The most relevant first
  • With candidates to review: Check which conditions each result satisfies
  • Intelligently filtered: Fewer irrelevant results

Interpreting results

Reasoned results:

  • Need review against your intent
  • Understand the context of use
  • Relate concepts intelligently
  • Prioritize relevance over quantity

Tip: Review the first candidates and check the details you need. If they do not fit, refine the query; their position does not guarantee that they meet all your criteria.

Best practices

Use it for complex searches

Reserve reasoned for:

  • Searches with multiple criteria
  • Abstract or emotional concepts
  • Cases where precision is critical
  • Searches with specific context

Be specific but natural

Write the way you'd describe the image to a designer:

  • ✅ "photographs of fashion products on a neutral background with professional lighting for a catalog"
  • ✅ "nature images that capture the serenity of dawn in the mountains"
  • ❌ "photos"
  • ❌ "jpg images of products"

Combine it with other methods

For maximum efficiency:

  1. Start with semantic for fast discovery
  2. Refine with reasoned if you need more precision
  3. Use basic if you remember the file name

Limitations and considerations

Processing time

Reasoned search:

  • Has variable timing depending on the query and service
  • Requires more processing resources
  • Offers candidates to compare for queries with several criteria

Tip: Use semantic first, and only use reasoned if you need more precision.

Very generic searches

For very simple searches:

  • It may be overkill, semantic may be enough
  • Compare its usefulness with the semantic results
  • Better to use semantic for general searches

Availability

Reasoned search:

  • May require that images be analyzed
  • Depends on external AI services
  • May have limits depending on your plan

Frequently asked questions

Semantic search finds images by similarity to your query. Reasoned search adds a second step: an AI model analyzes those candidates and reorders them based on the context and intent of the search. It can take longer than semantic search.

For searches with several criteria at once, abstract concepts such as a mood, or queries where the intended use matters, for example "luxury products on an elegant background for a premium campaign." For general or quick searches, semantic search is usually enough.

Click the search bar, select "Reasoned" mode, type your query and press Enter. The icon in the bar shows you're in that mode while the results are being processed.

Can you combine reasoned search with filters?

Yes, you can combine it with folder filters and the other available filters. Still review the top candidates: their position doesn't guarantee they meet all your criteria.

See also: semantic search.

Next steps

  1. Try reasoned search with a complex query
  2. Compare results with semantic search
  3. Experiment with abstract concepts and context
  4. Integrate it into your workflow for critical searches
  5. Combine it with other modes depending on the situation

To explore this topic further, see Basic search: finding files by name.