Feat: AI-powered natural language search for photos

[hk2166]

Describe the feature

Problem

Currently, PictoPy doesn't provide any special search feature.
There is no way to search using natural, conversational queries like
"show me photos from the beach" or "find pictures with my friends
from last year".

Proposed Solution

Implement a natural language search feature powered by AI that allows
users to query their photo library conversationally. This would involve:

• Generating text embeddings for each photo's metadata (tags, detected
  objects, faces, dates, locations)
• Storing embeddings in a vector index (e.g. FAISS)
• Accepting natural language queries from the user
• Converting queries to embeddings and retrieving the most semantically
  relevant photos

Additional AI Features (can be separate issues)

• Duplicate photo detection — identify near-identical images using
  embedding similarity and allow bulk removal
• Auto-generated albums — automatically group photos by event,
  location, or date using clustering

Why This Fits PictoPy

PictoPy already uses YOLO and FaceNet to extract rich metadata from
photos. This feature would make that metadata actually queryable in a
natural way, significantly improving the user experience without
compromising the privacy-first approach since all processing stays local.

Tech Stack Suggestion

• Python, FAISS, Sentence Transformers (all-MiniLM-L6-v2 for lightweight
  local embeddings)
• No external API calls — fully offline and privacy-preserving

Additional Notes

I'm interested in working on this feature and have prior experience
building RAG pipelines (FAISS + embeddings). Happy to discuss the
approach before starting implementation.

Add ScreenShots

Record

• I agree to follow this project's Code of Conduct
• I want to work on this issue

[github-actions]

👋 Thanks for opening this issue! The maintainers will review it shortly. Till then, do not open any PR. This is a strict rule we like to follow around here :)

[Arunodoy18]

Hi @hk2166,

This is a very exciting feature and aligns well with PictoPy’s privacy-first AI approach.

Before implementation, I’d like to clarify architecture to keep this scalable and maintainable.

Proposed phased approach:

Phase 1:

• Design embedding storage schema
• Evaluate local embedding model (e.g., all-MiniLM-L6-v2)
• Implement embedding generation pipeline (metadata → embedding)
• Persist embeddings locally

Phase 2:

• Integrate FAISS index
• Support query → embedding → semantic retrieval
• Add basic search UI

Phase 3 (optional extensions):

• Duplicate detection (embedding similarity)
• Auto album clustering

I would first like to propose a minimal technical design document for review before starting implementation.

Would that approach work?

[hk2166]

@rahulharpal1603 Can you review this issue, Does it works for the project or not. if yes please let me know