invoke-ai/InvokeAI · error · ValueError

{layer_key} not expected

Error message

{layer_key} not expected

What it means

lora_model_from_flux_control_state_dict maps each FLUX Control LoRA layer to an InvokeAI patch layer. After handling known layer kinds (lora A/B weights, bias, 'scale'), any layer whose state dict contains none of the expected entries triggers this ValueError. It's a strict-format guard for FLUX Control LoRA state dicts.

Source

Thrown at invokeai/backend/patches/lora_conversions/flux_control_lora_utils.py:84

            layers[prefixed_key] = FluxControlLoRALayer(
                layer_state_dict["lora_B.weight"],
                None,
                layer_state_dict["lora_A.weight"],
                None,
                layer_state_dict["lora_B.bias"],
            )
        elif all(k in layer_state_dict for k in ["lora_A.weight", "lora_B.bias", "lora_B.weight"]):
            layers[prefixed_key] = LoRALayer(
                layer_state_dict["lora_B.weight"],
                None,
                layer_state_dict["lora_A.weight"],
                None,
                layer_state_dict["lora_B.bias"],
            )
        elif "scale" in layer_state_dict:
            layers[prefixed_key] = SetParameterLayer("scale", layer_state_dict["scale"])
        else:
            raise ValueError(f"{layer_key} not expected")

    return ModelPatchRaw(layers=layers)

View on GitHub (pinned to 0b6a024f2f)

Solutions

1. Identify the offending layer_key and either remove it from the state dict (if it's a harmless extra tensor) before calling the converter.
2. Add a branch in flux_control_lora_utils.py mapping the new key type to the appropriate layer, then upstream the fix.
3. Confirm the LoRA is actually a FLUX Control LoRA and use the matching converter (kohya/diffusers/onetrainer/xlabs) otherwise.

for key, lsd in state_dict.items():
    if not ("lora_A" in lsd or "lora_B" in lsd or "scale" in lsd or "bias" in str(lsd.keys())):
        print(f"unexpected control-lora layer {key}: {sorted(lsd.keys())}")

def is_expected_control_layer(lsd: dict) -> bool:
    return any(k in lsd for k in ("lora_A.weight", "lora_B.weight", "scale"))

try:
    patch = lora_model_from_flux_control_state_dict(sd, model)
except ValueError as e:
    logger.error("FLUX Control LoRA conversion failed: %s", e)
    patch = None

When it happens

Trigger: Calling lora_model_from_flux_control_state_dict with a state dict containing a layer key whose layer_state_dict lacks lora_A/lora_B weights, bias, and 'scale' entries - e.g. an unexpected extra tensor like a norm alpha or a new param type added by a newer exporter.

Common situations: Newer FLUX Control LoRA checkpoints that add key types the converter doesn't know; truncated or manually edited safetensors files; mixing keys from a non-Control FLUX LoRA into this converter.

Related errors

• LoRA model is in unsupported FLUX format
• Unknown lora: {lora_key}!
• LoRA "{lora_key}" already applied to transformer.
• LoRA "{lora_key}" already applied to CLIP encoder.
• LoRA "{lora_key}" already applied to T5 encoder.

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/ea6e8228b9037b6b. Report an issue: GitHub.