invoke-ai/InvokeAI · error · NotAMatchError

state dict does not look like a main model

Error message

state dict does not look like a main model

What it means

Before accepting a main checkpoint config, InvokeAI checks the state dict against a heuristic set of keys expected in a main model (_has_main_keys). If none of the characteristic keys are present, the file is not treated as a main model and NotAMatchError is raised. Typically this means you scanned a VAE, LoRA, ControlNet, embedding, or an empty file rather than a full UNet checkpoint.

Source

Thrown at invokeai/backend/model_manager/configs/main.py:432

        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)


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

View on GitHub (pinned to 0b6a024f2f)

Solutions

1. Confirm the scanned file is the full checkpoint (contains model.diffusion_model.* UNet weights), not a VAE/LoRA/ControlNet.
2. Re-check directory structure and scan the intended file; fix path mixups.
3. Re-download if the file is truncated (very small file size).
4. Inspect keys: sd.keys() should include UNet entries before import.

Example fix

// check first
from safetensors.torch import load_file
sd = load_file('model.safetensors')
assert any(k.startswith('model.diffusion_model') for k in sd), 'not a main checkpoint'

sd = mod.load_state_dict()
if not any(k.startswith(('model.diffusion_model', 'double_blocks', 'down_blocks')) for k in sd):
    print('File does not look like a main model checkpoint')

def looks_like_main_model(sd: dict) -> bool:
    from invokeai.backend.model_manager.configs.main import _has_main_keys
    return _has_main_keys(sd)

try:
    cfg = probe_model(path)
except NotAMatchError as e:
    if 'main model' in str(e):
        print('Not a main checkpoint — classify as VAE/LoRA/ControlNet as appropriate')

When it happens

Trigger: from_model_on_disk → _validate_looks_like_main_model on a state dict failing the main-keys heuristic — VAE files, LoRA files, ControlNet state dicts, text-encoder dumps, or files with pruned/renamed keys.

Common situations: Batch-scanning a models directory and hitting non-checkpoint files; picking the wrong file in a multi-file repo; all-in-one downloads where UNet keys are nested under unexpected prefixes.

Related errors

• Unrecognized LLLite module name: '{name}'
• State dict contains no LLLite modules (no 'lllite_dit_blocks
• LLLite module '{name}' is missing key '{down_key}'
• Unexpected key: {k}
• missing keys after fp8 load: {missing[:10]}

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/1273ba13410ecd01. Report an issue: GitHub.