Show HN: I made an OSS tool to remove Sora 2 Watermark in less than 72h released (github.com)

An open-source project called Sweeta was released that can remove the SORA 2 watermark from generated videos — the author says it was built in under 72 hours. Sweeta combines automatic watermark detection with advanced inpainting (LaMA-based) to reconstruct masked regions, and exposes both a GUI and CLI with options like detection sensitivity, transparent-watermark output, and format conversion. The repo is Apache‑2.0 licensed, credits a heavily modified WatermarkRemover‑AI model, and includes install notes and dependency fixes; the author frames the release as a prompt for stronger, more obvious provenance controls rather than malicious use, and adds an educational/research-only disclaimer. For the AI/ML community this is significant because it demonstrates how fragile visible watermarking can be: within days, a readily-usable tool undermined a platform’s provenance mark, raising practical concerns for content authenticity, moderation, and arms-race dynamics between watermarking and removal. Key technical takeaways: Sweeta relies on existing inpainting stacks (iopaint/LaMA), requires a moderately capable machine (≥16 GB RAM, NVIDIA GPU with ~4 GB VRAM or Apple Silicon), and offers tunable detection parameters. The release underscores the need for more robust, possibly cryptographic or tamper-evident provenance mechanisms, plus community discussion about responsible disclosure and deployment.