# LivePortrait CUDA checkpoint notices The seven safetensors files are lossless conversions of the official human LivePortrait weights at KlingTeam/LivePortrait revision 82a4fa6735ca58432b6ce39301b4b9ee066dea47. Original spectral normalization weight_orig/u/v tensors are retained. Conversion does not train the model. The conversion receipt records strict architecture loading and tensor roundtrips. landmark.onnx is the unchanged LivePortrait 203-point asset from that same checkpoint revision, under its liveportrait directory. The pinned model card labels the repository MIT. Preserve the complete upstream LICENSE, including its separate InsightFace exception; this notice does not override those terms. face_detection_yunet_2023mar.onnx is unchanged from opencv/opencv_zoo revision 47534e27c9851bb1128ccc0102f1145e27f23f98. The associated MIT copyright and license are included as YUNET_LICENSE. This selected CUDA checkpoint contains no InsightFace detector or landmark weights. It uses YuNet detection with the LivePortrait 203-point landmark asset. These are prepared publication assets, not proof of a published distribution, installation acceptance, or quality validation. Immutable Hugging Face and ModelScope origins for the complete converted file set are still required.