Closed petercham closed 1 month ago
Since all models are placed in the intranet shared directory, could it be because the IPAdapter models can be configured with the IPAdapter model extension directory via ComfyUI's extra_model_paths.yaml, and the InsightFace models, which are not included by ComfyUI in folder_path. py so it can only be loaded locally?
Including the IPEmbeds model, can only be loaded from the Input directory, but in fact the IPEmbeds model is not as variable as the input image, it will be relatively targeted training saved.
It is suggested that both types of models can be set to be loaded in a subdirectory of the IPAdapter model directory, so that sharing can be accomplished through ComfyUI's extra_model_paths.yaml configuration.
can I see the full error message? This looks like a problem with insightface installation
can I see the full error message? This looks like a problem with insightface installation
It is indeed possible that the problem is due to a faulty installation of InsightFace, and that a fresh installation of ComfyUI and IPAdapter_plus was not followed by a specific installation of Insightface
you probably miss onnxruntime and/or onnxruntime-gpu
alternatively you may have an older version of insightface installed
In my case, running pip install insightface
was always installing insightface===0.2.1
I then ran pip install insightface==0.7.3
and that installed the latest version and then the problem went away
The InsightFace Loader node failed to load and reported the following error:
ComfyUI version is 72508a8(https://github.com/comfyanonymous/ComfyUI/commit/72508a8d19121e2814ea4dfbce8a5311f37dcd61) May 4, 2024. IPAdapter_plus node version is a336127 (https://github.com/cubiq/ComfyUI_IPAdapter_plus/commit/a336127797485aeaa255cd20ecfe64b7f3d52d75) May 2, 2024.
All fresh installations, no residual issues with older IPAdapter_plus node versions.
The same error is reported regardless of CPU, CUDA, or ROCM selection.