Face swapper model fails to load - inconsistent model filename in code

[Ranb972]

Bug Description

The face swapper model fails to load due to inconsistent model filename references in the code. The application searches for both inswapper_128.onnx and inswapper_128_fp16.onnx at different times, but the documentation instructs users to download inswapper_128_fp16.onnx.

Environment

• OS: Windows 11
• Deep-Live-Cam Version: 2.0.1c
• Python Version: 3.11.9
• Execution Provider: CUDA (CUDAExecutionProvider)
• GPU: NVIDIA GeForce RTX 4050 Laptop GPU
• CUDA Version: 12.8
• cuDNN Version: 8.9.7

Steps to Reproduce

1. Follow the installation instructions from README
2. Download the model files as instructed:
   • GFPGANv1.4.pth
   • inswapper_128_fp16.onnx (as per documentation)
3. Place files in models/ directory
4. Run: python run.py --execution-provider cuda
5. Select a source face and click "Live"

Expected Behavior

The face swapper model should load successfully and swap faces in real-time.

Actual Behavior

The application repeatedly fails to load the model with the error:

[DLC.FACE-SWAPPER] Loading face swapper model from: C:\Deep-Live-Cam\models\inswapper_128_fp16.onnx
[DLC.FACE-SWAPPER] Error loading face swapper model: model_file C:\Deep-Live-Cam\models\inswapper_128_fp16.onnx should exist
[DLC.FACE-SWAPPER] Face swapper model not loaded or failed to load. Skipping swap.

Sometimes it also searches for inswapper_128.onnx (without _fp16).

Console Output

Applied providers: ['CUDAExecutionProvider', 'CPUExecutionProvider']
[DLC.FACE-SWAPPER] Loading face swapper model from: C:\Deep-Live-Cam\models\inswapper_128_fp16.onnx
[DLC.FACE-SWAPPER] Error loading face swapper model: model_file C:\Deep-Live-Cam\models\inswapper_128_fp16.onnx should exist
[DLC.FACE-SWAPPER] Face swapper model not loaded or failed to load. Skipping swap.

Impact

• Face swapping functionality completely non-functional
• Only Face Enhancer works (with CUDA)
• Application shows FPS: 3-4 (very slow due to missing face swap model)
• No actual face replacement occurs - only blur/enhancement

Additional Information

• The file inswapper_128_fp16.onnx exists in the models/ directory
• File size: ~528 MB (downloaded from the official HuggingFace link)
• GFPGANv1.4.pth loads successfully
• CUDA is properly configured and working (confirmed via PyTorch and ONNX Runtime tests)

Possible Solution

The code should consistently reference the correct model filename. Either:

1. Update code to use inswapper_128_fp16.onnx everywhere, OR
2. Update documentation to instruct users to rename the file to inswapper_128.onnx

Verification

I verified that:

• ✅ PyTorch sees CUDA: torch.cuda.is_available() returns True
• ✅ ONNX Runtime has CUDA: ort.get_available_providers() includes CUDAExecutionProvider
• ✅ GPU is properly configured in Windows Graphics Settings
• ✅ All dependencies installed correctly
• ❌ Face swapper model fails to load

Thank you for this amazing project! Looking forward to seeing this resolved.