lllyasviel/Fooocus · error · ValueError

Wrong activation value in EqualLinear: {activation}Supported

Error message

Wrong activation value in EqualLinear: {activation}Supported ones are: ['fused_lrelu', None].

What it means

EqualLinear is the equalized-learning-rate linear layer used by Fooocus's face-restoration StyleGAN2 architecture (ldm_patched/pfn/architecture/face/stylegan2_arch.py). Its constructor only accepts activation=None or 'fused_lrelu'; any other string raises ValueError immediately at module construction. This is a hard configuration check, not a runtime data error.

Source

Thrown at ldm_patched/pfn/architecture/face/stylegan2_arch.py:168

            Supported: 'fused_lrelu', None. Default: None.
    """

    def __init__(
        self,
        in_channels,
        out_channels,
        bias=True,
        bias_init_val=0,
        lr_mul=1,
        activation=None,
    ):
        super(EqualLinear, self).__init__()
        self.in_channels = in_channels
        self.out_channels = out_channels
        self.lr_mul = lr_mul
        self.activation = activation
        if self.activation not in ["fused_lrelu", None]:
            raise ValueError(
                f"Wrong activation value in EqualLinear: {activation}"
                "Supported ones are: ['fused_lrelu', None]."
            )
        self.scale = (1 / math.sqrt(in_channels)) * lr_mul

        self.weight = nn.Parameter(torch.randn(out_channels, in_channels).div_(lr_mul))
        if bias:
            self.bias = nn.Parameter(torch.zeros(out_channels).fill_(bias_init_val))
        else:
            self.register_parameter("bias", None)

    def forward(self, x):
        if self.bias is None:
            bias = None
        else:
            bias = self.bias * self.lr_mul
        if self.activation == "fused_lrelu":
            out = F.linear(x, self.weight * self.scale)

View on GitHub (pinned to ae05379cc9)

Solutions

1. Set activation=None (plain linear, activation applied externally) or activation='fused_lrelu' in the EqualLinear call.
2. If you need a different activation, keep EqualLinear(activation=None) and apply nn.LeakyReLU/etc. on the output tensor in the parent module instead.
3. If the value comes from a checkpoint/config dict, inspect it (print the activation field) and remap legacy names to the supported set before construction.

Example fix

// before
EqualLinear(512, 512, activation='lrelu')
// after
EqualLinear(512, 512, activation=None)  # then apply F.leaky_relu(out) yourself

valid = {None, 'fused_lrelu'}
assert activation in valid, f'activation must be one of {valid}, got {activation!r}'

def is_equal_linear_activation(v) -> bool:
    return v is None or (isinstance(v, str) and v == 'fused_lrelu')

When it happens

Trigger: Instantiating EqualLinear(in_channels, out_channels, activation='relu') (or 'lrelu', 'gelu', etc.) directly, or building a StyleGAN2 generator/from a checkpoint/config that serializes an activation string other than 'fused_lrelu' or None.

Common situations: Porting a StyleGAN2 implementation that used a different activation name, editing the face-enhancement pipeline (experiments_face.py / pfn arch) to try a new activation, or deserializing a model config written for another repo where the same class accepts 'fused_lrelu' spelled differently.

Related errors

• Wrong sample mode {self.sample_mode}, supported ones are ['u
• Wrong activation value in EqualLinear: {activation}Supported
• Max depth of recursive function `tie_encoder_to_decoder` rea
• invalid distribution {distribution}
• network_name={network_name}

AI-assisted analysis of lllyasviel/Fooocus@ae05379cc9 (2026-08-15). Data as JSON: /api/errors/30e12c1c1426691f. Report an issue: GitHub.