deepfakes/faceswap · error · FaceswapError

Penalized Mask Loss has been selected but you have not chose

Error message

Penalized Mask Loss has been selected but you have not chosen a Mask to use. Please select a mask or disable Penalized Mask Loss.

What it means

Constructor-time validation in the training ModelBase: the config has penalized_mask_loss enabled but mask_type is 'none'. Penalized mask loss applies a penalty outside a mask, so it is meaningless without one; Faceswap refuses to start training rather than silently training without the penalty.

Source

Thrown at plugins/train/model/_base/model.py:67

        logger.debug(parse_class_init(locals()))
        # Input shape must be set within the plugin after initializing
        self.input_shape: tuple[int, ...] = ()
        """A `tuple` of `ints` defining the shape of the faces that the model takes as input. This
        should be overridden by model plugins in their :func:`__init__` function. If the input size
        is the same for both sides of the model, then this can be a single 3 dimensional `tuple`.
        If the inputs have different sizes for `"A"` and `"B"` this should be a `list` of 2 3
        dimensional shape `tuples`, 1 for each side respectively."""

        self.color_order: T.Literal["bgr", "rgb"] = "bgr"  # Override for image color channel order

        self._args = arguments
        self._is_predict = predict
        self._model: keras.Model | None = None

        cfg.load_config(config_file=arguments.config_file)

        if cfg.Loss.penalized_mask_loss() and cfg.Loss.mask_type() == "none":
            raise FaceswapError("Penalized Mask Loss has been selected but you have not chosen a "
                                "Mask to use. Please select a mask or disable Penalized Mask "
                                "Loss.")

        if cfg.Loss.learn_mask() and cfg.Loss.mask_type() == "none":
            raise FaceswapError("'Learn Mask' has been selected but you have not chosen a Mask to "
                                "use. Please select a mask or disable 'Learn Mask'.")

        self._mixed_precision = cfg.mixed_precision()
        self._io = IO(self, model_dir,
                      self._is_predict,
                      T.cast(T.Literal["never", "always", "exit"], cfg.Optimizer.save_optimizer()))
        self._check_multiple_models()

        self._state = State(model_dir,
                            self.name,
                            False if self._is_predict else self._args.no_logs)
        self._settings = Settings(self._args,
                                  self._mixed_precision,

View on GitHub (pinned to f530cb7508)

Solutions

1. Set Mask > mask_type in the training config to an actual mask that exists for your faces (see also error 40).
2. Or disable Loss > penalized_mask_loss if you want to train without masks.
3. Use the GUI config editor, which makes the incompatible combination visible when reviewing settings.

Example fix

# before (training config)
mask_type = none
penalized_mask_loss = True

# after
mask_type = vgg-clear
penalized_mask_loss = True
# or: penalized_mask_loss = False with mask_type = none

from lib.config import cfg
cfg.load_config(config_file=arguments.config_file)
assert not (cfg.Loss.penalized_mask_loss() and cfg.Loss.mask_type() == "none"), \
    "penalized_mask_loss requires a mask_type"

def mask_loss_config_valid(penalized: bool, mask_type: str) -> bool:
    return not (penalized and mask_type == "none")

When it happens

Trigger: Creating/initiating a training session where cfg.Loss.penalized_mask_loss() is True and cfg.Loss.mask_type() == 'none' in the model's training configuration.

Common situations: User disables the mask (to save VRAM or simplify) but forgets they earlier enabled penalized mask loss; hand-editing the loss section of the config; toggling settings independently in the GUI.

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
• 'Learn Mask' has been selected but you have not chosen a Mas
• Landmark based masks cannot be created for {self._landmark_t

AI-assisted analysis of deepfakes/faceswap@f530cb7508 (2026-08-15). Data as JSON: /api/errors/7c55f77a76eb07a7. Report an issue: GitHub.