You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Enabling the face correction option while using a GFPGAN model in combination with a Stable Diffusion upscaling model will then pass the SD model to the ESRGAN background upscaler, causing an error:
File "C:\Users\ssube\stabdiff\onnx-try-2\onnx-web\api\onnx_web\chain\base.py", line 93, in __call__
image = stage_pipe(ctx, stage_params, params, image,
File "C:\Users\ssube\stabdiff\onnx-try-2\onnx-web\api\onnx_web\chain\correct_gfpgan.py", line 72, in correct_gfpgan
gfpgan = load_gfpgan(ctx, upscale, upsampler=upsampler)
File "C:\Users\ssube\stabdiff\onnx-try-2\onnx-web\api\onnx_web\chain\correct_gfpgan.py", line 34, in load_gfpgan
upsampler = load_resrgan(ctx, upscale)
File "C:\Users\ssube\stabdiff\onnx-try-2\onnx-web\api\onnx_web\chain\upscale_resrgan.py", line 34, in load_resrgan
raise Exception('Real ESRGAN model not found at %s' % model_path)
Exception: Real ESRGAN model not found at ..\models\upscaling-stable-diffusion-x4.onnx
The error is complaining that the .onnx file does not exist, but SD pipelines are stored in a directory, and wouldn't work for the ESRGAN upscaler anyway.
The text was updated successfully, but these errors were encountered:
The Real ESRGAN upsampler for GFPGAN was something that I had unintentionally enabled, so I've removed that for now in favor of the upscale model/tab that is already available. That allows mixing SD upscaling with GFPGAN, but if you set face outscale > 1, forces GFPGAN to use Lanczos interpolation instead of an SR model.
Enabling the face correction option while using a GFPGAN model in combination with a Stable Diffusion upscaling model will then pass the SD model to the ESRGAN background upscaler, causing an error:
The error is complaining that the
.onnx
file does not exist, but SD pipelines are stored in a directory, and wouldn't work for the ESRGAN upscaler anyway.The text was updated successfully, but these errors were encountered: