WatermarkRemover-AI 1.0 (Omni)

WatermarkRemover-AI 1.0 (Omni)

Today we are releasing WatermarkRemover-AI 1.0, an open-source inference pipeline for automated watermark detection and removal.

This release stabilizes the underlying model architecture, shifting from traditional heuristics to a deterministic two-stage neural network approach. The pipeline is fully localized, removing any reliance on cloud compute or external APIs.

Architecture Highlights

Florence-2 Integration

We have integrated Microsoft's Florence-2-large foundation model for spatial detection. By leveraging its robust zero-shot object detection capabilities, the system can reliably isolate watermarks, burnt-in subtitles, and persistent logos regardless of the underlying video context.

Latent Inpainting Pipeline

For the reconstruction phase, we utilize the Large Mask Inpainting (LaMA) architecture. Once the spatial coordinates of the watermark are identified by Florence-2, LaMA generates context-aware pixel data to fill the masked regions. This ensures the output maintains temporal consistency and high structural integrity across video frames.

Automated Dependency Resolution

Setting up local ML environments is notoriously brittle. In 1.0, we have overhauled the setup scripts (setup.bat, setup.ps1, setup.sh) to automatically resolve and cache the required PyTorch runtimes, CUDA dependencies, and model weights on the first run. The regional proxy configurations have been removed to ensure a universal, zero-touch initialization process.

Hardware Acceleration

The inference engine now natively detects available hardware. It will automatically route tensor operations to CUDA-enabled GPUs when available, significantly reducing render times for high-resolution video pipelines. CPU fallback remains supported for standard image processing.

Installation

You can pull the 1.0 release directly from the repository.

git clone https://github.com/nihannihu/watermark-remover-ai.git
cd WatermarkRemover-AI
./setup.sh  # or setup.bat on Windows

For full documentation and advanced CLI flags, please refer to the updated README. We look forward to seeing what the community builds on top of this architecture.