lllyasviel/Fooocus · error · NotImplementedError

padding layer [{:s}] is not implemented

Error message

padding layer [{:s}] is not implemented

What it means

block.pad() creates the padding layer: 'reflect' -> ReflectionPad2d, 'replicate' -> ReplicationPad2d, and anything else raises NotImplementedError. Zero padding is deliberately not handled here (the comment says 'if padding is zero, do by conv layers') - conv blocks are expected to carry the padding themselves, so pad_type='zero' is also rejected.

Source

Thrown at ldm_patched/pfn/architecture/block.py:63

    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)
        )
    return layer


def get_valid_padding(kernel_size, dilation):
    kernel_size = kernel_size + (kernel_size - 1) * (dilation - 1)
    padding = (kernel_size - 1) // 2
    return padding


class ConcatBlock(nn.Module):
    # Concat the output of a submodule to its input
    def __init__(self, submodule):
        super(ConcatBlock, self).__init__()
        self.sub = submodule

    def forward(self, x):

View on GitHub (pinned to ae05379cc9)

Solutions

1. Use 'reflect' or 'replicate', or set padding=0 so the function returns None and lets conv layers handle zero padding
2. For 'zero' configs: drop the explicit pad layer and put the padding into the following nn.Conv2d(padding=...)

Example fix

# before
p = B.pad('zero', 1)  # -> NotImplementedError

# after: zero pad via the conv itself
conv = nn.Conv2d(64, 64, 3, 1, 1)  # padding=1 IS the zero padding
block = nn.Sequential(conv)

def sanitize_pad(pad_type, padding):
    if padding == 0:
        return None
    p = pad_type.lower()
    assert p in ('reflect', 'replicate'), f"pad_type must be reflect/replicate (zero padding belongs in the conv), got {pad_type!r}"
    return p

# Real-ESRGAN configs often say 'zero': translate instead of failing
if cfg.get('pad_type') == 'zero':
    cfg['pad_type'] = 'reflect'; cfg['padding'] = 0  # conv supplies zero pad

def is_supported_pad(pad_type: str) -> bool:
    return pad_type.lower() in ('reflect', 'replicate')

When it happens

Trigger: Calling pad('zero', 1), pad('symmetric', 2), or pad('', 3) - any pad_type outside {'reflect','replicate'} with a non-zero padding amount. Note: if padding == 0 the function returns None before the check, so the error only fires when actual padding is requested with an unsupported type.

Common situations: Real-ESRGAN configs commonly specify pad_type: zero - porting them into this vendored code without changing the value; configs requesting 'symmetric' padding available in other forks.

Related errors

• activation layer [{:s}] is not found
• normalization layer [{:s}] is not found
• scale {scale} is not supported. Supported scales: 2^n and 3.
• scale {scale} is not supported. Supported scales: 2^n and 3.
• Upsample mode [{self.upsampler}] is not found

AI-assisted analysis of lllyasviel/Fooocus@ae05379cc9 (2026-08-15). Data as JSON: /api/errors/bfe49fdcdda249ef. Report an issue: GitHub.