Project Overview
Sora2 Watermark Remover is a specialized application combining computer vision and deep learning models to detect and non-destructively remove watermarks or logos embedded in videos (such as those produced by recent AI models).
The main goal is to reconstruct the masked areas as faithfully as possible to the original video, without excessive blur.
Key Features
- Automatic/semi-automatic watermark detection: Identifies regions containing the unwanted logo or text.
- Advanced video inpainting: Temporal and spatial pixel reconstruction algorithms that smoothly fill the removed areas.
- Batch processing: Process several video files in a row.
- Quality preservation: Keeps the original frame rate (FPS) and resolution.
Technical Details
- Programming language: Python for all image processing and inference logic.
- Computer vision: OpenCV for video decoding, image matrix manipulation and post-processing.
- AI inference: Deep inpainting models that estimate background textures and motion behind watermarks.