deep-live-cam

Deep-Live-Cam is an open-source real-time face swap and video deepfake tool. Given a single source face image, it can replace faces in live webcam feeds or in pre-recorded images and videos, with output available almost instantly. The project is designed for the AI-generated media industry, including animation of custom characters, content creation, and similar creative work. The README includes an ethical use disclaimer, noting that users should obtain consent when using a real person's face, clearly label deepfake output when sharing, and follow applicable laws. The software claims a built-in check that prevents processing inappropriate media such as nudity, graphic content, or sensitive material. The developers state they may add watermarks or shut down the project if legally required. Key features highlighted in the repository include: - Live webcam face swapping with a preview that can be streamed through tools like OBS. - Image and video face swapping, with results saved to a directory named after the target video. - Mouth mask option to retain the original mouth movement. - Face mapping to apply different faces to multiple subjects simultaneously. - Support for many faces in one frame. - CLI mode via the -s/--source argument for scripted use. Manual installation requires Python 3.11-3.14, pip, git, ffmpeg, and on Windows Visual Studio 2022 runtimes. Two model files are downloaded from Hugging Face and placed in a models folder: gfpgan-1024.onnx and inswapper_128_fp16.onnx. The project recommends using a virtual environment. Instructions are provided for Windows, Linux, and macOS, including specific steps for Apple Silicon using Python 3.14 and the tkinter package. Hardware acceleration supports multiple execution providers: - CUDA for NVIDIA GPUs, using CUDA Toolkit 12.8 and cuDNN, with onnxruntime-gpu. - CoreML for Apple Silicon, using the official onnxruntime build. - CoreML legacy for older Apple devices via onnxruntime-coreml. - DirectML for Windows devices, including AMD GPUs. - OpenVINO for Intel hardware, with specific version pairings between onnxruntime-openvino and OpenVINO. Command-line arguments include source, target, output, frame processor selection, keep-fps, keep-audio, many-faces, map-faces, mouth-mask, video encoder and quality options, live mirror, live resizable, max memory, execution provider, execution threads, and version. The project also promotes a separate pre-built 'Ultimate' version with additional features, optimizations, and priority support, available through its official website. The repository credits several prior open-source projects, including insightface for models and library code, roop as the base author's earlier project, and various contributors for webcam support, multiple face handling, and platform optimizations. It also lists press coverage from outlets such as Ars Technica, CNN Brasil, Bloomberg Technoz, TrendMicro, and PetaPixel, as well as reactions from streamers.