Comfy-Org/ComfyUI · error · ValueError

Unknown normalization type: {norm_type}

Error message

Unknown normalization type: {norm_type}

What it means

Factory guard in the ACE depth-conv VAE (comfy/ldm/ace/vae/autoencoder_dc.py): get_normalization() only constructs batch_norm, group_norm, layer_norm, and rms_norm. Any other norm_type string in the model config raises ValueError because no module can be built for the checkpoint's architecture description.

Source

Thrown at comfy/ldm/ace/vae/autoencoder_dc.py:36

    def forward(self, x):
        x = super().forward(x)
        if self.elementwise_affine:
            if self.bias is not None:
                x = x + comfy.model_management.cast_to(self.bias, dtype=x.dtype, device=x.device)
        return x


def get_normalization(norm_type, num_features, num_groups=32, eps=1e-5):
    if norm_type == "batch_norm":
        return nn.BatchNorm2d(num_features)
    elif norm_type == "group_norm":
        return ops.GroupNorm(num_groups, num_features)
    elif norm_type == "layer_norm":
        return ops.LayerNorm(num_features)
    elif norm_type == "rms_norm":
        return RMSNorm(num_features, eps=eps, elementwise_affine=True, bias=True)
    else:
        raise ValueError(f"Unknown normalization type: {norm_type}")


def get_activation(activation_type):
    if activation_type == "relu":
        return nn.ReLU()
    elif activation_type == "relu6":
        return nn.ReLU6()
    elif activation_type == "silu":
        return nn.SiLU()
    elif activation_type == "leaky_relu":
        return nn.LeakyReLU(0.2)
    else:
        raise ValueError(f"Unknown activation type: {activation_type}")


class ResBlock(nn.Module):
    def __init__(
        self,

View on GitHub (pinned to 1c6d8d45b3)

Solutions

1. Open the checkpoint's config and check the norm_type fields; map them to one of batch_norm|group_norm|layer_norm|rms_norm
2. If the config uses a synonym (e.g. 'batchnorm'), correct it to the exact supported spelling
3. If the checkpoint genuinely needs an unsupported norm, update this fork to handle it in get_normalization before loading

Example fix

# before (checkpoint config)
"norm_type": "batchnorm"

# after
"norm_type": "batch_norm"

SUPPORTED_NORMS = {"batch_norm", "group_norm", "layer_norm", "rms_norm"}
assert all(nt in SUPPORTED_NORMS for nt in cfg['norm_type'] if isinstance(nt, str)) or cfg['norm_type'] in SUPPORTED_NORMS

def is_supported_norm(norm_type: str) -> bool:
    return norm_type in {"batch_norm", "group_norm", "layer_norm", "rms_norm"}

When it happens

Trigger: Building AutoencoderDC (or a block calling get_normalization) from a config whose norm_type is e.g. 'group_norm_32', 'BN', 'sync_batchnorm', or a typo like 'gruop_norm'. The value comes straight from the checkpoint's JSON config, not user node inputs.

Common situations: Loading a fine-tuned or community-modified ACE/DC-VAE checkpoint whose config was edited or came from a newer upstream that added a normalization type this code does not support; porting a diffusers AutoencoderDC config with renamed norm types.

Related errors

• Unknown activation type: {activation_type}
• Block with {block_type=} is not supported.
• Normalization {name} not found
• Normalization mode {self.qkv_norm_mode} not found, only supp
• Invalid normalization type: {normtype}

AI-assisted analysis of Comfy-Org/ComfyUI@1c6d8d45b3 (2026-08-14). Data as JSON: /api/errors/0f71c439627a1959. Report an issue: GitHub.