deepfakes/faceswap · error · FaceswapError

No weights were successfully loaded from your weights file

Error message

No weights were successfully loaded from your weights file: '{self._weights_file}'. Please check and try again.

What it means

Raised after the weight-transfer pass when loaded_ops == 0: the .keras weights file opened successfully, but not a single layer of the current model matched/successfully received weights from it. Faceswap treats total transfer failure as an error (partial mismatches only produce a warning with skipped_ops).

Solutions

Confirm the weights file was saved from the same model plugin and settings you are training now.
If you changed input size or other architecture-affecting settings, revert them to match the weights, or drop the weights file.
Check the log for the skipped-layers warning to see which layers failed and why (shape mismatch vs name mismatch).
If intentional (training from scratch), remove the load-weights option.

Example fix

# before: weights from a different model plugin
faceswap train ... -m mymodel -wm other_model_weights.keras

# after: either use matching weights
faceswap train ... -m mymodel -m same_arch_weights.keras
# or train fresh without -wm

Defensive patterns

Strategy: validation

Validate before calling

# Before training, sanity-check that the weights model shares layer names with your model:
import keras.models as km
w = km.load_model(weights_file, compile=False)
my = km.load_model(state_file, compile=False)
shared = {l.name for l in w.layers} & {l.name for l in my.layers}
assert shared, "weights share no layers with this model; transfer will fail"

Prevention

Only warm-start from weights saved by the same plugin and architecture-affecting settings (input size, etc.).
Read the skipped-layers warning after enabling load-weights; zero loaded layers means the setup is wrong.

When it happens

Trigger: Loading a weights file whose architecture shares no layers with the current model — e.g. warm-starting from a different model plugin or a model trained with incompatible input size/settings, so every _load_layer_weights call returns 0.

Common situations: Trying to transfer weights between different model types; input dimensions or plugin settings changed so layer names/shapes never match; using a weights file saved by a much older version.

Related errors

AI-assisted analysis of deepfakes/faceswap@f530cb7508 (2026-08-15). Data as JSON: /api/errors/9a42df79aef075a6. Report an issue: GitHub.

Appendix: source

Thrown at plugins/train/model/_base/io.py:494

                msg = f"Skipping layer {model_name} as not in "
                msg += "current_model." if not sub_model else f"weights '{self._weights_file}.'"
                logger.warning(msg)
                continue

            logger.info("Loading weights for layer '%s'", model_name)
            skipped_ops = 0
            loaded_ops = 0
            for layer in sub_model.layers:
                success = self._load_layer_weights(layer, sub_weights, model_name)
                if success == 0:
                    skipped_ops += 1
                elif success == 1:
                    loaded_ops += 1

        del weights_models

        if loaded_ops == 0:
            raise FaceswapError(f"No weights were successfully loaded from your weights file: "
                                f"'{self._weights_file}'. Please check and try again.")
        if skipped_ops > 0:
            logger.warning("%s weight(s) were unable to be loaded for your model. This is most "
                           "likely because the weights you are trying to load were trained with "
                           "different settings than you have set for your current model.",
                           skipped_ops)

    def _get_weights_model(self) -> list[k_models.Model]:
        """Obtain a list of all sub-models contained within the weights model.

        Returns
        -------
        List of all models contained within the .keras file

        Raises
        ------
        FaceswapError
            In the event of a failure to load the weights, or the weights belonging to a different

View on GitHub (pinned to f530cb7508)