deepfakes/faceswap · error · FaceswapError

Load weights selected, but the path '{weights_file}' does no

Error message

Load weights selected, but the path '{weights_file}' does not exist. Please check and try again.

What it means

Raised while validating the 'load weights' input in the training model IO: the user selected a weights file, but its path does not exist (the subsequent branch rejects paths that are not .keras files). It is a pre-flight check that runs before any weights are actually read.

Source

Thrown at plugins/train/model/_base/io.py:424

        Returns
        -------
        The full path to a weights file
        """
        if not weights_file:
            logger.debug("[Weights] No weights file selected.")
            return None

        msg = ""
        if not os.path.exists(weights_file):
            msg = f"Load weights selected, but the path '{weights_file}' does not exist."
        elif not os.path.splitext(weights_file)[-1].lower() == ".keras":
            msg = (f"Load weights selected, but the path '{weights_file}' is not a valid Keras "
                   f"model (.keras) file.")

        if msg:
            msg += " Please check and try again."
            raise FaceswapError(msg)

        logger.verbose("Using weights file: %s", weights_file)  # type:ignore
        return weights_file

    def freeze(self) -> None:
        """If freeze has been selected in the cli arguments, then freeze those models indicated
        in the plugin's configuration. """
        # Blanket unfreeze layers, as checking the value of :attr:`layer.trainable` appears to
        # return ``True`` even when the weights have been frozen
        for layer in get_all_sub_models(self._model):
            layer.trainable = True

        if not self._do_freeze:
            logger.debug("[Weights] Freeze weights deselected. Not freezing")
            return

        for layer in get_all_sub_models(self._model):
            if layer.name in self._freeze_layers:

View on GitHub (pinned to f530cb7508)

Solutions

1. Correct the path so it points at the existing .keras weights file (use an absolute path).
2. Confirm the extension is '.keras' — a valid file with the wrong extension fails the next check in the same function.
3. If you no longer want to warm-start, clear the load-weights setting.

Example fix

# before
faceswap train ... -wm weights/model.h5

# after
faceswap train ... -wm /home/user/weights/model.keras

import os
assert not weights_path or (os.path.exists(weights_path)
                             and os.path.splitext(weights_path)[-1].lower() == ".keras"), \
    "weights path must exist and end in .keras"

def is_valid_weights_file(path: str | None) -> bool:
    return path is None or (os.path.isfile(path) and path.lower().endswith(".keras"))

When it happens

Trigger: Launching training with the load-weights option pointing at a non-existent path — typo, file moved/deleted, or relative path resolved from the wrong working directory.

Common situations: Passing --load-weights / -wm paths by hand after moving files; scripts running with a different cwd; placeholder paths left in launch commands.

Related errors

• You have selected the mask type '{mask_type}' but at least o
• '{method}' is not a valid clipping method. Select from {list
• '{name}' is not a valid optimizer. Select from {list(_OPTIMI
• No weights were successfully loaded from your weights file:
• Penalized Mask Loss has been selected but you have not chose

AI-assisted analysis of deepfakes/faceswap@f530cb7508 (2026-08-15). Data as JSON: /api/errors/13f917ee381889b8. Report an issue: GitHub.