I am having an error with library conflicts (PyTorch vs. ONNXRuntime) on new GPUs

[binhandeptrai-design]

Hello,

I am experiencing a major library conflict when trying to run Deep-Live-Cam on modern hardware.

The Problem: Two Runtimes

The core issue is that this project requires two different AI libraries to run at the same time:

1. PyTorch: Used for the "Face Enhancer" (GFPGANv1.4.pth).
2. onnxruntime-gpu: Used for the "Face Swapper" (inswapper_128_fp16.onnx).

The Consequence: Hardware Incompatibility

This dependency on both libraries creates a critical conflict on new hardware, like my NVIDIA GeForce RTX 5060 Ti.

My GPU has a new CUDA capability (sm_120), but the PyTorch version required by this project (e.g., cu128 or cu118) is too old and does not support this new architecture. This makes the "Face Enhancer" feature unusable and throws a UserWarning.

Proposed Solution: Switch to a Single Library (ONNX)

My proposed solution is to remove the PyTorch dependency completely. This would solve all the library conflicts and hardware incompatibility issues.

This can be done by converting the GFPGANv1.4.pth model into the .onnx format.

I use another similar application (waixingrenai_8.02_50X) which does exactly this. Its models folder contains gfpgan_1.4.onnx and codeformer.onnx, and it runs perfectly on my hardware because it only uses the onnxruntime-gpu library.

By making this change, the project would become much more stable, easier to install, and compatible with new hardware.

Thank you.

waixingrenai_8.02_50X:

models: