deepfakes/faceswap · error · FaceswapError

You have requested to train with the

Error message

You have requested to train with the '{self.name}' plugin, but a model file for the '{multiple_models[0]}' plugin already exists in the folder '{self.io.model_dir}'.\nPlease select a different model folder.

What it means

FaceswapError raised by ModelBase._check_multiple_models() when training is requested with one plugin (e.g. 'original') but the model folder already contains a state/model file belonging to a different single plugin. Faceswap state files are plugin-specific and incompatible across model architectures, so it refuses to mix them. This is a guard against silently corrupting an existing model.

Solutions

  1. Create a new, empty model folder and pass it with -m so the new plugin starts fresh
  2. Or set -t to the plugin named in the error message (multiple_models[0]) to continue training the existing model
  3. Or move the existing model files out of the folder if the folder is meant to be reused

Example fix

# before
python faceswap.py train -t original -m /models/shared
# error: folder holds 'dlight' model

# after
python faceswap.py train -t original -m /models/original_new
# or continue the existing model:
python faceswap.py train -t dlight -m /models/shared
Defensive patterns

Strategy: validation

Validate before calling

import os, glob
plugin = "original"  # the trainer you will launch
files = glob.glob(os.path.join(model_dir, "*_state.*")) + glob.glob(os.path.join(model_dir, "*.h5"))
foreign = [f for f in files if os.path.basename(f).split("_")[0] not in (plugin,)]
if foreign:
    raise SystemExit(f"Model folder holds foreign plugin files: {foreign}")

Try / catch

from lib.exceptions import FaceswapError
try:
    trainer = Trainer(model_dir, args)
except FaceswapError as err:
    if "plugin already exists" in str(err):
        print("Pick a fresh model folder or the matching -t plugin")
    raise

Prevention

When it happens

Trigger: Passing -m/--model-dir pointing at a folder that holds e.g. dlight_state.h5 while launching train.py with -t original (or any mismatched trainer plugin). Triggered on every training start that performs the model-folder consistency check.

Common situations: Reusing the same folder for a new experiment with a different model plugin, downloading someone else's model and training with the wrong plugin name, or a typo in the -t trainer argument.

Related errors

AI-assisted analysis of deepfakes/faceswap@f530cb7508 (2026-08-15). Data as JSON: /api/errors/0ade7761d8b81424. Report an issue: GitHub.

Appendix: source

Thrown at plugins/train/model/_base/model.py:188

        FaceswapError
            If multiple model files, or models for a different plugin from that requested exists
            within the model folder
        """
        multiple_models = self._io.multiple_models_in_folder
        if multiple_models is None:
            logger.debug("Contents of model folder are valid")
            return

        if len(multiple_models) == 1:
            msg = (f"You have requested to train with the '{self.name}' plugin, but a model file "
                   f"for the '{multiple_models[0]}' plugin already exists in the folder "
                   f"'{self.io.model_dir}'.\nPlease select a different model folder.")
        else:
            p_types = "', '".join(multiple_models)
            msg = (f"There are multiple plugin types ('{p_types}') stored in the model folder '"
                   f"{self.io.model_dir}'. This is not supported.\nPlease split the model files "
                   "into their own folders before proceeding")
        raise FaceswapError(msg)

    def build(self) -> None:
        """Build the model and assign to :attr:`model`.

        Within the defined strategy scope, either builds the model from scratch or loads an
        existing model if one exists.

        If running inference, then the model is built only for the required side to perform the
        swap function, otherwise  the model is then compiled with the optimizer and chosen
        loss function(s).

        Finally, a model summary is outputted to the logger at verbose level.
        """
        is_summary = hasattr(self._args, "summary") and self._args.summary
        if self._io.model_exists:
            model = self.io.load()
            if self._is_predict:
                inference = Inference(model, self._args.swap_model)

View on GitHub (pinned to f530cb7508)