2.7-ultimate

The largest update in Deep‑Live‑Cam’s history is here. Version 2.7 adds over 35,000 lines of new code — a full rewrite focused on speed, reliability, and creative power.

🔥 Performance
- Integrated TensorRT acceleration for NVIDIA GPUs (up to 3× faster, no quality loss)
- Removed heavy dependencies (PyTorch, TensorFlow, InsightFace, GFPGAN) → smaller install, faster startup, lower memory use
- Face enhancer rebuilt fully on ONNX
- Batched multi‑face inference (swap many faces at once)
- Async face detection (camera never stalls)
- Smart caching + optimized image handling, masks, and rendering
- Tuned for smooth performance even on lower‑end GPUs

🎬 New Features
Flux Live — edit faces with text prompts
- RTX Upscaler — GPU sharpening for crisp frames
- Real‑time frame interpolation for smoother motion
- New Reswapper face model
- Target face selection & mapping

🎨 Quality Improvements
- Seamless blending (Poisson + feathering)
- Fixed lip‑mask symmetry, ghosting, and edge artifacts
- Better color matching between source and target
- Sharper, flicker‑free exports
- More accurate face alignment with detailed landmarks

🖥️ Interface & Experience
- Brand‑new PyQt6 UI — modern, fast, clean
- Classic View option for legacy fans
- Full theming system with custom themes
- 5‑language support: English, Arabic, French, Japanese, Chinese
- Auto window projection & resize for streaming

🛡️ Reliability & Device Support
- Smart installer auto‑detects GPU (NVIDIA incl. RTX 50‑series, AMD, Apple Silicon, enterprise GPUs)
- Video export runs in its own process → no more end‑stage crashes
- Fixed multi‑GPU AMD (DirectML) targeting
- Improved Apple Silicon acceleration
- Fixed startup/camera crashes across systems
- Clearer error messages with direct support links

[https://deeplivecam.net/index.php/quickstart](<img width="1877" height="461" alt="image" src="https://github.com/user-attachments/assets/3b60d6ba-c92e-479d-9471-9dc46be921c7" />)

Downloadable only at https://deeplivecam.net/index.php/quickstart