hacksider/deep-live-cam
real time face swap and one-click video deepfake with only a single image
How deep-live-cam is put together
Deep-Live-Cam is a desktop real-time face-swap / deepfake application written in Python with a PySide6 GUI. It captures webcam or video frames via OpenCV, detects and analyses faces with insightface/ONNX models (inswapper_128 for swapping, GFPGAN/GPEN for enhancement), applies per-frame processors (swapper, enhancers, masking), and writes output images/videos or streams live. Frame processors are pluggable modules under modules/processors/frame with a shared core; models are lazily downloaded to a local 'models' directory at first run. Execution providers (CUDA, CoreML, DirectML, OpenVINO) are selected via CLI flags in run.py.
Languages
Frameworks
Infrastructure
Major components
modules.processors.frame.face_swapper
Core face-swap processor: runs the inswapper ONNX model, performs affine paste-back and Poisson/seamless blending, with caching and Apple Silicon optimizations.
modules.face_analyser
Detects one or many faces in frames using insightface models and provides source/target face selection utilities.
modules.ui
PySide6-based GUI: source/target image selection, camera picker, live preview, processor options, and localization-aware widgets.
modules.core / modules.run
Orchestrates end-to-end runs: pre-checks, frame iteration over images/videos/webcam, invoking enabled frame processors, and writing output.
modules.capturer / modules.video_capture
Reads video files and enumerates/captures webcam streams (with cv2_enumerate_cameras and pygrabber on Windows).
modules.model_downloader
Downloads ONNX model variants (e.g., inswapper_128.onnx/fp16) into the local models directory when missing.
modules.gpu_processing / modules.platform_info
Provides GPU-accelerated image ops (blur, sharpen, resize) and platform/execution-provider configuration including OpenVINO settings.
Frame processors (face_enhancer, gpen256/512, face_masking)
Optional post-processing steps: GFPGAN/GPEN face restoration and face masking applied after swapping.
Setup got easier with automatic model downloads and refreshed instructions.
5 weeksWeek of 2026-08-17
Week of 2026-08-17latest
Setup got easier with automatic model downloads and refreshed instructions.
Week of 2026-08-10
Two fixes landed, improving face mapping and Linux camera detection.
Week of 2026-08-03
Documentation clarified which OpenVINO versions work with which runtime.
Week of 2026-07-27
Dependencies were upgraded, including a newer GPU acceleration library.
Week of 2026-07-20
A video seeking bug was fixed to prevent out-of-range jumps.
Over the past five weeks, the deep-live-cam project focused on polishing reliability and easing setup for everyday users. Fixes addressed video frame seeking, face-mapping frame extraction, and Linux camera device paths, while documentation and dependency updates clarified GPU/OpenVINO setup. The most recent week added automatic model downloads to make first-time installation smoother.
Week by week
2026-08-17Setup got easier with automatic model downloads and refreshed instructions.latest2 changes
Feature
Automatic model downloads
The app can now download some required AI models on its own, so users no longer have to fetch them manually.
Docs
Readme update
The project's main instruction page was updated to reflect the latest guidance.
2026-08-10Two fixes landed, improving face mapping and Linux camera detection.2 changes
Fix
Face mapping frame extraction fix
Fixed a problem where the fallback path for extracting frames when mapping faces could fail.
Fix
Linux camera device path fix
Corrected how the app finds camera devices on Linux so webcams are detected properly.
2026-08-03Documentation clarified which OpenVINO versions work with which runtime.1 change
Docs
OpenVINO version pairing note
Added a note explaining which versions of onnxruntime-openvino and OpenVINO should be used together to avoid compatibility issues.
2026-07-27Dependencies were upgraded, including a newer GPU acceleration library.2 changes
Chore
Requirements update
The list of software the project depends on was refreshed.
Chore
onnxruntime-gpu upgrade to 1.26.0
The GPU acceleration component was bumped to version 1.26.0 for better performance and compatibility.
2026-07-20A video seeking bug was fixed to prevent out-of-range jumps.1 change
Fix
Clamped video frame seek index
Seeking within a video is now limited to valid positions, preventing errors from jumping beyond the start or end of the clip.
Dependencies and code review
Dependency advisories
Security Watch
No known advisories across 11 scanned dependencies.
No known advisories in the scanned dependencies.
Code review
No concrete code-level issues confirmed.
Get this report every week for your repos.
GitZoid learns each repo, reports what changed, and flags what needs attention. One flat price for the whole team.
$19 a month, flat · First 10 outputs free · No card required