Fix the issue where CUDA version less than 12. x may cause running on CPU

[JI4JUN]

When I run this node in the CUDA11.8 environment, there is a situation where the CPU is running at full load while the GPU is not, resulting in extremely slow running speed of the video face swapping workflow.

I found that the installation of onnxrruntime-gpu 1.12.1, which is not compatible with the current environment, was caused by running the install.by script.
elif not is_installed(ort,"1.16.1",False): run_pip(ort, "-U")

After I specified the installation of version 1.16.1, I could clearly feel that the GPU was involved in the computation
run_pip(ort, "onnxruntime-gpu==1.16.1")

By the way, the nsfw check also takes a lot of time due to the default use of CPU.
predict = pipeline("image-classification", model=model_path)

[chandujr]

Was this a CUDA-specific issue? I have a similar problem with onnxruntime-rocm. The swapping is done by CPU and only the face restoration is done by GPU.

[xb191799981]

我也碰到这个问题了，发现ReActor一直用CPU处理图片， 22秒的原视频换脸， 需要消耗23分钟才能完成。

我本地显卡是4090-48G，CUDA版本信息：
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Fri_Feb_21_20:42:46_Pacific_Standard_Time_2025
Cuda compilation tools, release 12.8, V12.8.93
Build cuda_12.8.r12.8/compiler.35583870_0

Name: onnxruntime-gpu
Version: 1.22.0
Summary: ONNX Runtime is a runtime accelerator for Machine Learning models
Home-page: https://onnxruntime.ai
Author: Microsoft Corporation
Author-email: [email protected]
License: MIT License
Location: f:\ai\comfyui-aki-v1.4-self\python\lib\site-packages
Requires: coloredlogs, flatbuffers, numpy, packaging, protobuf, sympy
Required-by: #N/A