AUTOMATIC1111/stable-diffusion-webui · error · ValueError

No GFPGAN model found

Error message

No GFPGAN model found

What it means

Raised by gfpgan_model.load_model after it iterated every candidate .pth file (found via modelloader with ext_filter ['.pth']) in the GFPGAN model directory and none had 'GFPGAN' in the filename basename, or none existed at all. It means no GFPGAN face-restoration weights are available where the loader expects them.

Source

Thrown at modules/gfpgan_model.py:44

    def get_device(self):
        return devices.device_gfpgan

    def load_net(self) -> torch.Module:
        for model_path in modelloader.load_models(
            model_path=self.model_path,
            model_url=model_url,
            command_path=self.model_path,
            download_name=model_download_name,
            ext_filter=['.pth'],
        ):
            if 'GFPGAN' in os.path.basename(model_path):
                return modelloader.load_spandrel_model(
                    model_path,
                    device=self.get_device(),
                    expected_architecture='GFPGAN',
                ).model
        raise ValueError("No GFPGAN model found")

    def restore(self, np_image):
        def restore_face(cropped_face_t):
            assert self.net is not None
            return self.net(cropped_face_t, return_rgb=False)[0]

        return self.restore_with_helper(np_image, restore_face)


def gfpgan_fix_faces(np_image):
    if gfpgan_face_restorer:
        return gfpgan_face_restorer.restore(np_image)
    logger.warning("GFPGAN face restorer not set up")
    return np_image


def setup_model(dirname: str) -> None:
    global gfpgan_face_restorer

View on GitHub (pinned to 82a973c043)

Solutions

1. Put a valid GFPGAN weight file (e.g. GFPGANv1.4.pth) into the model directory shown by the loader (typically <webui-root>/models/GFPGAN) keeping 'GFPGAN' in the filename and a .pth extension.
2. Or let the UI download it: open the Extras/face-restore tab and trigger the model download so modelloader fetches it from the registered model_url.
3. Alternatively switch face restorer to CodeFormer if GFPGAN weights are unavailable.

Example fix

# before: file renamed so basename check fails
models/GFPGAN/gfpgan_v14.pth

# after: keep 'GFPGAN' and .pth in the name
models/GFPGAN/GFPGANv1.4.pth

import os, glob

def gfpgan_available(model_dir):
    return any('GFPGAN' in os.path.basename(p)
               for p in glob.glob(os.path.join(model_dir, '**', '*.pth'), recursive=True))

if shared.opts.face_restoration_model == 'GFPGAN' and not gfpgan_available('models/GFPGAN'):
    print('GFPGAN weights missing; face restore will fail')

try:
    restorer = gfpgan_model.GfpganRestorer(...)
except ValueError as e:
    if 'No GFPGAN model found' in str(e):
        download_gfpgan_weights()  # then retry once
    else:
        raise

When it happens

Trigger: Enabling CodeFormer/GFPGAN face restore (restore_faces) with the GFPGAN restorer selected, while the models/GFPGAN directory is empty or contains no file whose basename contains 'GFPGAN' with a .pth extension.

Common situations: Fresh install where the automatic download failed or was skipped, the model file was renamed (e.g. to GFPGANv1.4.pth.backup), or it was placed in the wrong directory.

Related errors

• No codeformer model found
• When merging inpainting model with a normal one, A must be t
• When merging instruct-pix2pix model with a normal one, A mus
• Model file {path} not found
• hypernetwork uses an unsupported activation function: {activ

AI-assisted analysis of AUTOMATIC1111/stable-diffusion-webui@82a973c043 (2026-08-14). Data as JSON: /api/errors/dd41478c837c0edc. Report an issue: GitHub.