deepfakes/faceswap · error · FaceswapError

Too many identities: {num_identities}. Max: {len(identities)

Error message

Too many identities: {num_identities}. Max: {len(identities)}

What it means

Identity labels are built from single characters: A-Z (26), then 0-9 (10), then a-z (26), capping at 62 concurrent identities. If the number of detected identities in the dataset exceeds 62, this FaceswapError aborts label assignment. It reflects a hard design limit of the labelling scheme, not a resource issue.

Source

Thrown at lib/training/data/data_set.py:72

    ----------
    index
        The index of the current label
    num_identities
        The number of identities that belong to the label set
    next_identity
        ``True`` to return the next identity for the given index. Default: ``False``

    Returns
    -------
    The current or next label. Labels go A-Z,0-9,a-z
    """
    identities = [chr(i) for i in range(65, 65 + 26)]
    if num_identities > len(identities):
        identities += [chr(i) for i in range(48, 48 + 10)]
    if num_identities > len(identities):
        identities += [chr(i) for i in range(97, 97 + 26)]
    if num_identities > len(identities):
        raise FaceswapError(f"Too many identities: {num_identities}. Max: {len(identities)}")
    identities = identities[:num_identities]
    index = index % num_identities
    if not next_identity:
        return identities[index]
    index += 1 if index + 1 < num_identities else -index
    return identities[index]


def get_sorted_images(folder: str) -> list[str]:
    """For the given folder return the sorted list of potential training images

    Parameters
    ----------
    folder
        The folder containing faceswap training images

    Returns
    -------

View on GitHub (pinned to f530cb7508)

Solutions

Reduce the number of identities: raise the clustering distance/threshold so nearby faces merge.
Split the dataset into subsets of <=62 identities and process separately.
Filter out low-face-count identity clusters (noise) before labelling.

Example fix

# identity plugin config
# before
identity_threshold = 0.3  # over-splits -> 80 identities -> FaceswapError
# after
identity_threshold = 0.6  # merges clusters under 62 identities

Defensive patterns

Strategy: validation

Validate before calling

MAX_IDENTITIES = 62
assert num_identities <= MAX_IDENTITIES, \
    f'{num_identities} identities exceeds label limit {MAX_IDENTITIES}; raise threshold or split data'

Type guard

def within_identity_limit(n: int) -> bool:
    return isinstance(n, int) and 0 < n <= 62

Try / catch

try:
    label = get_identity_label(num_identities, index)
except FaceswapError as err:
    if 'Too many identities' in str(err):
        raise SystemExit('raise identity_threshold or split the dataset')
    raise

Prevention

Tune the identity clustering threshold so clusters stay under 62.
Prune single-face noise clusters before labelling.
Split very large multi-person datasets into per-group runs.

When it happens

Trigger: Running identity-aware training/sorting on a dataset clustering into more than 62 distinct identities; identity threshold set low enough that noise creates many small clusters.

Common situations: Large multi-person datasets (crowd footage, celebrity sets) in identity plugin usage; overly-sensitive clustering producing spurious extra identities.

Related errors

AI-assisted analysis of deepfakes/faceswap@f530cb7508 (2026-08-15). Data as JSON: /api/errors/d2af322650b1848d. Report an issue: GitHub.