Face swap / deepfake workflow on RX 9070 XT under Arch Linux?
Hi everyone,
I'm trying to build a reliable deepfake / face swap workflow on Arch Linux using an AMD Radeon RX 9070 XT (tried ROCm 6.4, 7.0).
So far I haven't been able to get a working pipeline. ReActor doesn't work correctly on my system. The GPU is detected by PyTorch (torch.cuda.is_available() == True), but face swapping either falls back to the CPU or produces broken results
My goal is simple:
* use a **source image** (the face that should be swapped in),
* use a **target video** (a different person),
* process the video frame by frame,
* replace the face on every frame,
* then reassemble the processed frames back into the final video.
Has anyone with an RX 9070 XT or another RDNA4 GPU successfully done this in ComfyUI?
Which nodes or tools are you using?
* ReActor
* FaceFusion
* SimSwap
* InstantID
* PuLID
* Something else?
I'm mainly looking for a workflow that works reliably with AMD ROCm on Arch Linux.
Any recommendations, working workflows, or setup tips would be greatly appreciated. Thanks!