invoke-ai/InvokeAI · error · NotAMatchError

unrecognized unet in_channels {in_channels} for base '{base}

Error message

unrecognized unet in_channels {in_channels} for base '{base}'

What it means

After reading the UNet's first-conv in_channels, only 4 (Normal), 5 (Depth, SD2 only), and 9 (Inpaint) are recognized. Any other channel count — or 5 on a base other than SD2 — raises NotAMatchError because InvokeAI has no variant mapping for it.

Source

Thrown at invokeai/backend/model_manager/configs/main.py:426

        state_dict = mod.load_state_dict()
        key_name = "model.diffusion_model.input_blocks.0.0.weight"

        if key_name not in state_dict:
            raise NotAMatchError("unable to determine model variant from state dict")

        in_channels = state_dict["model.diffusion_model.input_blocks.0.0.weight"].shape[1]

        match in_channels:
            case 4:
                return ModelVariantType.Normal
            case 5:
                # Only SD2 has a depth variant
                assert base is BaseModelType.StableDiffusion2, f"unexpected unet in_channels 5 for base '{base}'"
                return ModelVariantType.Depth
            case 9:
                return ModelVariantType.Inpaint
            case _:
                raise NotAMatchError(f"unrecognized unet in_channels {in_channels} for base '{base}'")

    @classmethod
    def _validate_looks_like_main_model(cls, mod: ModelOnDisk) -> None:
        has_main_model_keys = _has_main_keys(mod.load_state_dict())
        if not has_main_model_keys:
            raise NotAMatchError("state dict does not look like a main model")


class Main_Checkpoint_SD1_Config(Main_SD_Checkpoint_Config_Base, Config_Base):
    base: Literal[BaseModelType.StableDiffusion1] = Field(default=BaseModelType.StableDiffusion1)


class Main_Checkpoint_SD2_Config(Main_SD_Checkpoint_Config_Base, Config_Base):
    base: Literal[BaseModelType.StableDiffusion2] = Field(default=BaseModelType.StableDiffusion2)


class Main_Checkpoint_SDXL_Config(Main_SD_Checkpoint_Config_Base, Config_Base):
    base: Literal[BaseModelType.StableDiffusionXL] = Field(default=BaseModelType.StableDiffusionXL)

View on GitHub (pinned to 0b6a024f2f)

Solutions

Verify the file is a main UNet checkpoint, not a ControlNet/adapter — import ControlNets via their own model type.
If it is a custom-channel fine-tune, InvokeAI cannot classify it; use the upstream repo tooling instead or patch the input conv back to 4 channels if it is a leftover.
Check the base resolution: in_channels==5 is only valid for SD2 Depth; ensure the correct base was detected.
Update InvokeAI in case support for newer channel layouts was added.

Example fix

// sanity check before import
sd = load_file('model.safetensors')
ch = sd['model.diffusion_model.input_blocks.0.0.weight'].shape[1]
assert ch in (4, 5, 9), f'unsupported in_channels {ch}'

Defensive patterns

Strategy: validation

Validate before calling

sd = load_file('model.safetensors')
ch = sd['model.diffusion_model.input_blocks.0.0.weight'].shape[1]
if ch not in (4, 5, 9):
    print(f'in_channels={ch}: likely a ControlNet/adapter or custom fine-tune, not a main model')

Type guard

def has_supported_variant(sd: dict) -> bool:
    w = sd.get('model.diffusion_model.input_blocks.0.0.weight')
    return w is not None and w.shape[1] in (4, 5, 9)

Try / catch

try:
    cfg = probe_model(path)
except NotAMatchError as e:
    if 'in_channels' in str(e):
        print('Import as its proper model type (e.g. ControlNet) instead')

Prevention

Import ControlNets/T2I adapters as their own model type
Avoid custom-channel fine-tunes unless you know how to serve them
Check in_channels early when triaging unknown checkpoints

When it happens

Trigger: from_model_on_disk → _get_variant_or_raise with in_channels not in {4,5,9}, or in_channels==5 while the resolved base is not StableDiffusion2 (the assert fires first with a different message only if assertion checks pass differently; the match fallthrough raises this error otherwise).

Common situations: ControlNet or T2I-Adapter weights (extra conditioning channels) mistaken for main checkpoints; custom fine-tunes with modified input convs (e.g. grayscale or 8-channel editors); updated models adding channels (e.g. 8 for image-conditioned edit models).

Related errors

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/add3ff91a8dd3a47. Report an issue: GitHub.