deepfakes/faceswap · error · FaceswapError

You do not have enough GPU memory available to run detection

Error message

You do not have enough GPU memory available to run detection at the selected batch size. Youcan try a number of things:
1) Close any other application that is using your GPU (web browsers are particularly bad for this).
2) Try again. Sometimes this can be a transient issue when you are close to VRAM capacity.
3) Lower the batch size (the amount of images fed into the model) by editing the plugin settings (GUI: Settings > Configure extract settings, CLI: Edit the file faceswap/config/extract.ini).
4) Use lighter weight plugins.
5) Enable fewer plugins.

What it means

FaceswapError wrapping a framework OutOfMemoryError raised inside plugin.process during detection/inference. The handler pads undersized batches to the plugin's configured batch size before predicting, so VRAM consumption tracks the configured batch size; when it exceeds available GPU memory this user-facing OOM_MESSAGE is raised with remediation steps.

Source

Thrown at lib/infer/handler.py:180

        The prediction from the model

        Raises
        ------
        FaceswapError
            If an OOM occurs
        """
        feed_size = feed.shape[0]
        is_padded = self.do_compile and feed_size < self.plugin.batch_size
        batch_feed = feed
        if is_padded:  # Prevent model re-compile on undersized batch
            batch_feed = np.empty((self.plugin.batch_size, *feed.shape[1:]), dtype=feed.dtype)
            logger.debug("[%s.process] Padding undersized batch of shape %s to %s",
                         self.plugin.name, feed.shape, batch_feed.shape)
            batch_feed[:feed_size] = feed
        try:
            retval = self.plugin.process(batch_feed)
        except OutOfMemoryError as err:
            raise FaceswapError(OOM_MESSAGE) from err
        if is_padded and retval.dtype == "object":
            out = np.empty(retval.shape, dtype="object")
            out[:] = [x[:feed_size] for x in retval]
            retval = out
        elif is_padded:
            retval = retval[:feed_size]
        return retval

    def _format_images(self, images: npt.NDArray[np.uint8]) -> np.ndarray:
        """Format the incoming UINT8 0-255 images to the format specified by the plugin

        Parameters
        ----------
        images
            The batch of UINT8 images to format

        Returns
        -------

View on GitHub (pinned to f530cb7508)

Solutions

1. Lower the batch size in plugin settings (GUI: Settings > Configure extract settings; CLI: edit faceswap/config/extract.ini).
2. Close other GPU consumers (browsers, other ML jobs) and retry — transient when near capacity.
3. Switch to lighter-weight or fewer plugins.
4. If persistent, use a smaller model or a GPU with more VRAM.

Example fix

# faceswap/config/extract.ini
# before
detect.batch_size = 64
# after
detect.batch_size = 8

# Before a long run, verify free VRAM vs batch size heuristic
import subprocess
free_mib = int(subprocess.run(
    ['nvidia-smi', '--query-gpu=memory.free', '--format=csv,noheader,nounits'],
    capture_output=True, text=True).stdout.split()[0])
batches_per_gpu = max(1, min(configured_batch, free_mib // 150))  # ~150MB/batch-item heuristic

try:
    detect(batch=batch_size)
except FaceswapError as err:
    if 'GPU memory' in str(err):
        batch_size //= 2
        detect(batch=batch_size)
    else:
        raise

When it happens

Trigger: Running detect/extract plugins with a batch size too large for the GPU; another process (browser, another training job) occupying VRAM; VRAM fragmentation when close to capacity.

Common situations: New users defaulting to high batch sizes on small GPUs (2-4GB); running extraction while a model trains; driver/browser compositing eating several hundred MB.

Related errors

• You do not have enough GPU memory available to train the sel
• Hex color codes should start with a '#' and be 6 characters
• An unhandled exception occurred initializing the device via
• There was an error reading from the Nvidia Machine Learning

AI-assisted analysis of deepfakes/faceswap@f530cb7508 (2026-08-15). Data as JSON: /api/errors/1d5ac0244b9e0de3. Report an issue: GitHub.