Honest limitation: this runs a heuristic (edge + texture pattern analysis) entirely in your browser, not a trained model, so it won’t be pixel-perfect on every photo the way a dedicated ML-based service is. It’s tuned to skip broad soft regions (shadows, gradients) and favor sparse, thin, high-contrast marks. Use the sensitivity slider or manual touch-up if it under- or over-selects. Only remove watermarks from images you have rights to edit.