[IMPORTANT ] Check your Reactor uses GPU -- not CPU!!

[Buumcode]

On market now we have two versions of ONNXRUNTIME -- for CPU and for GPU (onnxruntime and onnxruntime-gpu).

And here is big MISTAKE (big problem) -- BOTH versions of packages after installing became ONNXRUNTIME <---- get the same name!!

So, if you have installed in your system ONNXRUNTIME <----- your Reactor uses your CPU, NOT GPU.
(GPU two times faster than CPU -- which especially important in image-to-video face swapping).

Example 1 -- if you have installed BOTH onnxruntime and onnxruntime-gpu -- you still uses CPU:

pip list

onnx 1.16.1
onnxruntime 1.20.1
onnxruntime-gpu 1.20.1
opencv-python 4.10.0.84

Example 2 -- here is good config (my today's config) - only onnxruntime-gpu is installed = you uses your GPU:

pip list

onnx 1.17.0
onnxruntime-gpu 1.22.0
open_clip_torch 2.31.0

---- What to do ----
If your PIP LIST show you - you have installed onnxruntime -- you should completely uninstall them. But IMPORTANT - you should uninstall BOTH your onnxruntime and onnxruntime-gpu first!! And only after uninstalling BOTH versions of onnxruntime -- you need to install ONLY one onnxruntime-gpu.

Here is your steps:

pip uninstall onnxruntime
pip uninstall onnxruntime-gpu

pip install onnxruntime-gpu

Now, if you do it right -- your PIP LIST show you only onnxruntime-gpu is installed -- and you run your Reactor on GPU! 🌌🚀

[Buumcode]

Here is ZIP for testing on CPU-GPU.
How to use:
unpack to custom_nodes, as ~:\ComfyUI\custom_nodes\comfyui-reactor-node
make your test and look in console
***In every node you have line 'Device' -- type in it (small letters): 'auto', or 'cpu', or 'gpu'

--- Console Results (example, in short) ---
-Using CPU session for Hyperswap <--- you use CPU
-Using GPU session for Hyperswap <--- you use GPU
-inswapper_128.onnx with providers: ['CPUExecutionProvider'] <--- you use CPU
-inswapper_128.onnx with providers: ['CUDAExecutionProvider', 'CPUExecutionProvider'] <--- you use GPU
-Face Restore using device: cuda <------- GPU
-Face Restore using device: cpu <----- CPU

Reactor_v0.6.1-b2_CPU-GPU_face-restore_DEBUG.zip <---- 👀

[Elic8763]

I dont know how to use pip uninstall onnxruntime

[Buumcode]

I dont know how to use pip uninstall onnxruntime

'pip uninstall onnxruntime' - it's command in cmd.exe window.
Ask ChatGPT what to do. (Ask in your native language) - https://gpt-chatbot.ru/gpt-5

[erasmion]

is this still the case or fixed in the latest version?

[Buumcode]

is this still the case or fixed in the latest version?

It's still the case - because it's not problem with Reactor, but problem with 'onnxruntime' vs 'onnxruntime-gpu'.
You should activate your Virtual Environment (.venv) - if you use it - and check pip list for installed packages. In your pip list ONLY 'onnxruntime-gpu' need to be show as installed. Here is my example (today check) - I've don't installed 'onnxruntime':

omegaconf 2.3.0
onnx 1.17.0
onnx2torch 1.5.15
onnxruntime-gpu 1.23.2
open_clip_torch 2.31.0

[erasmion]

thanks for the quick help reply - did all you suggested with a portable installation

python.exe -m pip uninstall onnxruntime
then
python.exe -m pip uninstall onnxruntime-gpu
and finally
python.exe -m pip install onnxruntime-gpu

it worked, though i have not noticed much speed improvement yet

onnx 1.20.0
onnxruntime-gpu 1.23.2

Thank you!

[Buumcode]

it worked, though i have not noticed much speed improvement yet

Image2image face swap is pretty fast even on CPU. But Face Restore has much difference. Especially for image2video face swap.

Btw, some updating (ComfyUI, or custom nodes) - can install 'onnxruntime' - and you can see a bit slower work. Always check 'requirements.txt' before install something (to prevent install 'onnxruntime') - or check pip list to recognize them and uninstall (with 3 steps described above).

Happy face swapping!

[erasmion]

will do, thanks again Buumcode! : )

[kammeliyaDo]

works thanks you.

[poisas]

onnx 1.20.0
onnxruntime 1.23.2
onnxruntime-gpu 1.23.2
mine started to run on gpu today without any reason :)

[hum-ma]

Glad I found this discussion, thanks Buumcode. First I had version 1.23.2 of both packages installed but nothing could be detected due to this error:
[E:onnxruntime:, sequential_executor.cc:572 ExecuteKernel] Non-zero status code returned while running Relu node. Name:'Relu_2' Status Message: CUDA error cudaErrorNoKernelImageForDevice:no kernel image is available for execution on the device

So it actually was trying to use the GPU but failing probably because they have dropped support for good old GTX 970. I tried downgrading the packages and the error went away but at the same time ReActor started running on CPU only.

After finding this topic, I uninstalled the onnxruntime package, then tried different versions of onnxruntime-gpu again, and finally found that 1.20.2 is the latest that supports this GPU. Now it finally works without errors and fast too.