ProPainter is an open-source video inpainting system from ICCV 2023 that removes objects and watermarks from videos and fills the gaps with temporally consistent content. It improves on earlier methods by combining flow-guided pixel propagation with a transformer that hallucinates the parts no frame ever saw. The project has about 7,000 stars on GitHub and was updated in 2025.
Video inpainting is harder than image inpainting: the fill must look right in every frame and move naturally between them. ProPainter tackles this in stages โ it completes the optical flow field first, uses it to propagate real pixels from neighboring frames into the masked region, and then runs a mask-guided sparse transformer that generates content for areas that remain unknown, with far less memory than dense attention. The result is cleaner object removal, more reliable watermark and subtitle erasure, and plausible video completion and outpainting, even on longer clips. The repository ships inference code and pretrained weights, an interactive demo, and support for user-provided masks, and it has become a reference implementation that many later video-editing tools build on. It runs on a local GPU with PyTorch; as a research release it favors quality over speed, so expect offline processing rather than real-time editing.
ProPainter is free and open source for research and personal use. You provide your own GPU compute to run it.
ProPainter is for video editors, researchers and tool builders who need high-quality object removal or watermark cleanup in footage and are comfortable running a PyTorch model locally.