A beginner's guide to the Sora2-Watermark-Remover model by Uglyrobot on Replicate

Source: Dev.to

Model overview

The sora2-watermark-remover model, developed by uglyrobot, provides a solution for removing watermarks from videos generated by OpenAI’s Sora 2 system. Unlike general video‑editing tools such as propainter, which handle broader object‑removal tasks, this model specifically targets the watermark patterns found in Sora‑2 and Sora‑2‑Pro generated content. It combines a custom‑trained YOLOv11s detection system with the LAMA inpainting algorithm to identify and cleanly remove watermarks while preserving video quality.

Model inputs and outputs

Inputs

• video – The video file containing Sora 2 watermarks that need removal.
• output_filename – Optional custom filename for the cleaned video output.

Outputs

• Cleaned video file – A processed video with watermarks removed, maintaining the original video quality and content.

Capabilities

The model excels at detecting and removing watermarks from Sora 2‑generated videos, delivering a clean output without noticeable artifacts.

Read the full guide to Sora2‑Watermark‑Remover