Comfy-Org/ComfyUI · error · NotImplementedError

spatial and temporal padding modes must be equal

Error message

spatial and temporal padding modes must be equal

What it means

Raised by make_conv_nd in conv_nd_factory.py when spatial_padding_mode differs from temporal_padding_mode while causal=False. In the non-causal 3D path a single padding_mode applies to the whole Conv3d, so spatial and temporal modes must agree; mixed modes are only expressible in the causal DualConv3d path.

Source

Thrown at comfy/ldm/lightricks/vae/conv_nd_factory.py:24

import comfy.ops
ops = comfy.ops.disable_weight_init

def make_conv_nd(
    dims: Union[int, Tuple[int, int]],
    in_channels: int,
    out_channels: int,
    kernel_size: int,
    stride=1,
    padding=0,
    dilation=1,
    groups=1,
    bias=True,
    causal=False,
    spatial_padding_mode="zeros",
    temporal_padding_mode="zeros",
):
    if not (spatial_padding_mode == temporal_padding_mode or causal):
        raise NotImplementedError("spatial and temporal padding modes must be equal")
    if dims == 2:
        return ops.Conv2d(
            in_channels=in_channels,
            out_channels=out_channels,
            kernel_size=kernel_size,
            stride=stride,
            padding=padding,
            dilation=dilation,
            groups=groups,
            bias=bias,
            padding_mode=spatial_padding_mode,
        )
    elif dims == 3:
        if causal:
            return CausalConv3d(
                in_channels=in_channels,
                out_channels=out_channels,
                kernel_size=kernel_size,

View on GitHub (pinned to 1c6d8d45b3)

Solutions

1. Set spatial_padding_mode == temporal_padding_mode (usually both 'zeros')
2. Or pass causal=True so the DualConv3d path, which supports separate modes, is selected
3. Check both padding-mode fields in the VAE config for accidental divergence

Example fix

# before
conv = make_conv_nd(3, c_in, c_out, 3, causal=False, spatial_padding_mode="replicate", temporal_padding_mode="zeros")

# after
conv = make_conv_nd(3, c_in, c_out, 3, causal=False, spatial_padding_mode="zeros", temporal_padding_mode="zeros")

if not causal and spatial_padding_mode != temporal_padding_mode:
    raise ValueError("non-causal convs require spatial_padding_mode == temporal_padding_mode")

When it happens

Trigger: Calling make_conv_nd(dims=3, padding_mode-ish args with spatial_padding_mode='replicate', temporal_padding_mode='zeros', causal=False), or constructing a non-causal video VAE whose config sets different spatial/temporal padding modes.

Common situations: Porting configs that mix replicate spatial padding with zero temporal padding; forgetting to set causal=True when experimenting with padding modes.

Related errors

• unsupported dimensions: {dims}
• unknown block: {block_name}
• Invalid latent_log_var: {latent_log_var}
• Invalid input shape: {sample.shape}
• unknown layer: {block_name}

AI-assisted analysis of Comfy-Org/ComfyUI@1c6d8d45b3 (2026-08-14). Data as JSON: /api/errors/4f444d57d311f015. Report an issue: GitHub.