Infinite Image Browsing (IIB)

To minimize costs and processing time, the system uses incremental updates and caching for embeddings and titles.

1. Embedding Cache (image_embedding)

Embeddings are stored in the image_embedding table keyed by image_id. An image is skipped during updates if it has the same model, same text_hash, and an existing vec is present.

Re-vectorization logic: text_hash = sha256(f"{normalize_version}:{prompt_text}")

• prompt_text: Extracted and optionally normalized text.
• normalize_version: A code-derived fingerprint of the normalization rules used.

To force a rebuild: Pass force=true to build_iib_output_embeddings or force_embed=true to cluster_iib_output_job_start.

2. Title Cache (topic_title_cache)

Titles and keywords are stored in the topic_title_cache table keyed by cluster_hash.

Cache Hit Rule: When use_title_cache=true and force_title=false, titles are reused. The cluster_hash includes:

• Member image IDs (sorted)
• Embedding model, threshold, min_cluster_size
• title_model, output lang
• Normalization fingerprint (normalize_version) and mode

To force regeneration: Pass force_title=true.

To use IIB with an AI assistant, follow these two steps:

1. Start the IIB Service

Run the IIB server on a specific port:

python app.py --port <port>

2. Interact with the AI

Once the service is running, you can give the AI assistant natural language instructions for various tasks:

Search & Filtering:

• "Find all images with 'sunset' in the prompt"
• "Show images generated using model X"
• "Find images marked as 'favorite'"
• "Search for high-quality landscape images"

Tag Management:

• "Tag these images as 'high-quality'"
• "Remove the 'test' tag from all images"
• "Add a 'portrait' tag to images containing people"

File Organization:

• "Organize my downloads folder by theme"
• "Move all landscape images to a separate folder"

Information Retrieval:

• "Show the generation parameters for this image"
• "What prompts were used for these images?"
• "Compare the settings of these two images"

To automatically organize images into a destination folder using AI, follow this three-step process:

1. Start organization: Call POST /db/organize_files_start. You must provide folder_paths, dest_folder, a similarity threshold, lang, and an action (e.g., move). This returns a job_id.
2. Poll status: Use GET /db/organize_files_status?job_id=<job_id> to check the progress. Wait until the status is completed.
3. Confirm execution: Once completed, call POST /db/organize_files_confirm with the job_id to finalize the changes.

# 1. Start AI organization
curl -X POST http://127.0.0.1:<port>/infinite_image_browsing/db/organize_files_start \
  -H "Content-Type: application/json" \
  -d '{
    "folder_paths": ["/Downloads/ai-images"],
    "dest_folder": "/Pictures/organized",
    "threshold": 0.85,
    "lang": "en",
    "action": "move"
  }'
# Returns: {"job_id": "abc123"}

# 2. Poll for completion
curl "http://127.0.0.1:<port>/infinite_image_browsing/db/organize_files_status?job_id=abc123"
# Wait until status is "completed"

# 3. Confirm and execute
curl -X POST http://127.0.0.1:<port>/infinite_image_browsing/db/organize_files_confirm \
  -H "Content-Type: application/json" \
  -d '{"job_id": "abc123"}'

To include a new directory in the IIB index, follow these steps:

1. Add the path: Use POST /db/extra_paths with the path and desired types (e.g., ["scanned"]).
2. Update image data: Trigger an index update for that specific folder using POST /db/update_image_data with the path.
3. Verify: Search for images in that folder using POST /db/search_by_substr with the folder_paths parameter to ensure they are indexed.

# 1. Add path
curl -X POST http://127.0.0.1:<port>/infinite_image_browsing/db/extra_paths \
  -H "Content-Type: application/json" \
  -d '{"path": "/new/image/folder", "types": ["scanned"]}'

# 2. Trigger index update
curl -X POST http://127.0.0.1:<port>/infinite_image_browsing/db/update_image_data \
  -H "Content-Type: application/json" \
  -d '{"path": "/new/image/folder"}'

# 3. Verify images are indexed
curl -X POST http://127.0.0.1:<port>/infinite_image_browsing/db/search_by_substr \
  -H "Content-Type: application/json" \
  -d '{"surstr": "", "folder_paths": ["/new/image/folder"], "cursor": "", "size": 10}'