What causes "normalization layer [{:s}] is not found" in lllyasviel/Fooocus?

Pass norm_type='batch' or 'instance', or None/False at the conv_block level to skip normalization entirely

lllyasviel/Fooocus · error · NotImplementedError

normalization layer [{:s}] is not found

Error message

normalization layer [{:s}] is not found

What it means

block.norm() is the normalization factory: it builds BatchNorm2d for 'batch' and InstanceNorm2d for 'instance'; every other norm_type raises NotImplementedError. There is no 'none'/null option here - conv_block callers must skip normalization at a higher level rather than pass a string like 'none'.

Source

Thrown at ldm_patched/pfn/architecture/block.py:46

        layer = nn.LeakyReLU(neg_slope, inplace)
    elif act_type == "prelu":
        layer = nn.PReLU(num_parameters=n_prelu, init=neg_slope)
    else:
        raise NotImplementedError(
            "activation layer [{:s}] is not found".format(act_type)
        )
    return layer


def norm(norm_type: str, nc: int):
    # helper selecting normalization layer
    norm_type = norm_type.lower()
    if norm_type == "batch":
        layer = nn.BatchNorm2d(nc, affine=True)
    elif norm_type == "instance":
        layer = nn.InstanceNorm2d(nc, affine=False)
    else:
        raise NotImplementedError(
            "normalization layer [{:s}] is not found".format(norm_type)
        )
    return layer


def pad(pad_type: str, padding):
    # helper selecting padding layer
    # if padding is 'zero', do by conv layers
    pad_type = pad_type.lower()
    if padding == 0:
        return None
    if pad_type == "reflect":
        layer = nn.ReflectionPad2d(padding)
    elif pad_type == "replicate":
        layer = nn.ReplicationPad2d(padding)
    else:
        raise NotImplementedError(
            "padding layer [{:s}] is not implemented".format(pad_type)

View on GitHub (pinned to ae05379cc9)

Solutions

Pass norm_type='batch' or 'instance', or None/False at the conv_block level to skip normalization entirely
Sanitize config values: map 'none'/''/null to None before they reach norm()

Example fix

# before
layer = B.norm('none', 64)  # -> NotImplementedError

# after: skip normalization at the caller level
conv = B.conv_block(64, 64, norm_type=None)  # norm() never called

Defensive patterns

Strategy: validation

Validate before calling

def sanitize_norm(norm_type):
    if norm_type is None:
        return None
    n = norm_type.lower()
    if n in ('', 'none', 'null', 'false'):
        return None  # caller must then SKIP the norm layer
    assert n in ('batch', 'instance'), f'norm must be batch/instance/None, got {norm_type!r}'
    return n

conv = B.conv_block(64, 64, norm_type=sanitize_norm(cfg.get('norm')))

Type guard

def is_supported_norm(norm_type) -> bool:
    return norm_type is None or (isinstance(norm_type, str) and norm_type.lower() in ('batch', 'instance'))

Prevention

Use None/False to skip normalization; 'none' as a string is NOT accepted by norm()
Normalize config values: map 'none'/''/null to None before model construction

When it happens

Trigger: Calling norm('none', nc), norm('group', nc), or norm('batch0') - i.e. any norm_type besides 'batch'/'instance'. Typically triggered by a conv_block(act, norm_type='none') call pattern copied from code that treats 'none' as a valid sentinel.

Common situations: Configs from other BasicSR forks where norm_type: null / 'none' is written out explicitly; generated model params where the norm field is defaulted to a string instead of None; typos ('Batch', 'instanceNorm').

Related errors

AI-assisted analysis of lllyasviel/Fooocus@ae05379cc9 (2026-08-15). Data as JSON: /api/errors/f9b75a116e052838. Report an issue: GitHub.