On-product overlay (OPO) is an important indicator of device yield. In this work, we show that stressed thin films used in semiconductor manufacturing can be an important contributor to OPO at multiple length scales. Depending on the stress level, film thickness, and the mask design, the overlay impact can be a few nanometers for the exposure of the next lithography layer. A predictive compact model based on pattern density is developed to accurately predict this overlay impact. The model is then verified using short-loop dual damascene wafers with stress split. The predictive model opens a new opportunity for model-based mask correction during optical proximity correction to increase the overlay margin for subsequent lithography exposures.