|
|
---
|
|
|
title: Prism Content Moderation
|
|
|
emoji: 🛡️
|
|
|
colorFrom: blue
|
|
|
colorTo: purple
|
|
|
sdk: docker
|
|
|
app_port: 7860
|
|
|
pinned: false
|
|
|
license: mit
|
|
|
---
|
|
|
|
|
|
# Prism: AI-Powered Content Moderation
|
|
|
|
|
|
Prism is an automated content moderation pipeline that uses advanced AI models to audit images for compliance, safety, and quality.
|
|
|
|
|
|
## Features
|
|
|
- **Text Analysis:** OCR (EasyOCR) + VLM (InternVL) to check taglines, CTAs, and Terms & Conditions.
|
|
|
- **Visual Checks:** Detects ribbons, gestures, and assesses image quality.
|
|
|
- **Content Safety:** Flags offensive, risky (gambling/trading), or illegal content using CLIP and text analysis.
|
|
|
|
|
|
## Models Used
|
|
|
- **InternVL2_5-1B-MPO:** Multimodal LLM for detailed image understanding.
|
|
|
- **EasyOCR:** For robust text extraction.
|
|
|
- **CLIP (ViT-B/32):** For zero-shot image classification (theme detection).
|
|
|
|
|
|
## Usage
|
|
|
Upload an image to get a detailed compliance report.
|
|
|
|
