deepfakes/faceswap · error · FaceswapError

' is not compatible with your version of Keras. The minimum…

Error message

{arch}' is not compatible with your version of Keras. The minimum version required is {keras_min} whilst you have version {keras_ver} installed.

What it means

FaceswapError raised by Phaze-A._validate_encoder_architecture() when the chosen encoder architecture exists but your installed Keras version is below its keras_min requirement. Each entry in _MODEL_MAPPING declares a minimum Keras version; newer architectures often need newer Keras APIs. The error reports both the required minimum and your installed version.

Solutions

1. Upgrade Keras in the Faceswap environment to at least the version named in the message (pip install -U keras)
2. Or switch enc_architecture to an older encoder whose keras_min fits your installed version
3. Preferably let Faceswap's own requirements/installer manage the dependency set to avoid mismatches

Example fix

# before
pip show keras   # e.g. 2.15, arch requires 3.0

# after
pip install -U "keras>=3.0"
# or choose an older arch in phaze_a_config.json:
"enc_architecture": "efficientnet_b0"

from lib.utils import get_keras_version  # faceswap helper
from plugins.train.model.phaze_a import _MODEL_MAPPING
arch = _MODEL_MAPPING[cfg.enc_architecture()]
if get_keras_version() < arch.keras_min:
    raise SystemExit(f"Keras {arch.keras_min}+ required for {cfg.enc_architecture()}")

When it happens

Trigger: Selecting a modern encoder (e.g. a ViT or newer EfficientNet variant) while running an older Keras than its keras_min. get_keras_version() < model.keras_min at validation time.

Common situations: Pinned/downgraded Keras for another project, stale conda/venv environment, or Faceswap updated without refreshing its dependencies.

Related errors

• An unhandled exception occurred initializing the device via…
• ' ' is not a valid choice for encoder architecture. Choose…
• Clip network could not be found in
• Dataclass params should be a subset of dictionary keys
• Error loading weights file

AI-assisted analysis of deepfakes/faceswap@f530cb7508 (2026-08-15). Data as JSON: /api/errors/d7340fb5c5204444. Report an issue: GitHub.

Appendix: source

Thrown at plugins/train/model/phaze_a.py:336

        logger.debug("Encoder input set to: %s", retval)
        return retval

    def _validate_encoder_architecture(self) -> None:
        """ Validate that the requested architecture is a valid choice for the running system
        configuration.

        If the selection is not valid, an error is logged and system exits.
        """
        arch = cfg.enc_architecture()
        model = _MODEL_MAPPING.get(arch)
        if not model:
            raise FaceswapError(f"'{arch}' is not a valid choice for encoder architecture. Choose "
                                f"one of {list(_MODEL_MAPPING.keys())}.")

        keras_ver = get_keras_version()
        keras_min = model.keras_min
        if keras_ver < keras_min:
            raise FaceswapError(f"{arch}' is not compatible with your version of Keras. The "
                                f"minimum version required is {keras_min} whilst you have version "
                                f"{keras_ver} installed.")

    def build_model(self, inputs: list[KerasTensor]) -> keras.models.Model:
        """ Create the model's structure.

        Parameters
        ----------
        inputs: list[:class:`keras.KerasTensor`]
            A list of input tensors for the model. This will be a list of 2 tensors of
            shape :attr:`input_shape`, the first for side "a", the second for side "b".

        Returns
        -------
        :class:`keras.models.Model`
            The generated model
        """
        # Create sub-Models

View on GitHub (pinned to f530cb7508)