SemantiXel is a lightweight and modern web-based interface for performing semantic search on image datasets using CLIP and sentence embeddings. It enables intelligent retrieval of images based on text queries, image similarity, or embedded textual content, all in an elegant UI built for clarity and speed.
✨ Designed for creators, researchers, and developers to explore semantic media understanding with ease.
Get more familiar with Semantixel by knowing its purpose and what it offers:
For a detailed technical overview, architecture, setup instructions, and advanced usage, see the docs/ directory. It contains:
- System architecture and workflow
- Model and embedding details
- Data pipeline, search logic, UI/API, deployment, and more
- A glossary of key terms
Refer to these docs for in-depth understanding and implementation guidance.
-
Create a virtual environment:
conda create -n semantixel python=3.11 -y conda activate semantixel -
Install dependencies:
pip install -r requirements.txt -
Configure Settings:
python settings.py -
Run: (Creates Index + Runs Server + Launches UI)
python main.py
- 🔍 Text-to-Image Search using CLIP (
openai/clip-vit-base-patch32) - 🖼️ Image-to-Image Similarity Search via vision embeddings
- 📝 Embedded Text Search for documents and OCR content
- 🎛️ Customizable
thresholdandtop-Kranking - 💻 Fast, responsive UI with a clean white theme
- 🧠 Powered by HuggingFace Transformers & Doctr OCR
- 📂 Supports directory-level image indexing
- Retrieve screenshots showing "Apple Intelligence" in YouTube thumbnails
- Find similar photos or memes from your collection
- Detect specific phrases or embedded text in image-based documents
- Build your own AI-powered personal visual library