invoke-ai/InvokeAI · error · Exception

Unknown lora: {lora.lora.key}!

Error message

Unknown lora: {lora.lora.key}!

What it means

During FLUX LoRA collection, the invocation verifies each LoRA key exists in the model manager via context.models.exists(). If the key is absent it raises Exception('Unknown lora: ...'). This means the graph references a LoRA model record that no longer exists in the InvokeAI model registry.

Source

Thrown at invokeai/app/invocations/flux_lora_loader.py:167

        added_loras: list[str] = []

        if self.transformer is not None:
            output.transformer = self.transformer.model_copy(deep=True)

        if self.clip is not None:
            output.clip = self.clip.model_copy(deep=True)

        if self.t5_encoder is not None:
            output.t5_encoder = self.t5_encoder.model_copy(deep=True)

        for lora in loras:
            if lora is None:
                continue
            if lora.lora.key in added_loras:
                continue

            if not context.models.exists(lora.lora.key):
                raise Exception(f"Unknown lora: {lora.lora.key}!")

            if lora.lora.base is not BaseModelType.Flux:
                raise ValueError(
                    f"LoRA '{lora.lora.key}' is for {lora.lora.base.value if lora.lora.base else 'unknown'} models, "
                    "not FLUX models. Ensure you are using a FLUX compatible LoRA."
                )

            added_loras.append(lora.lora.key)

            if self.transformer is not None and output.transformer is not None:
                output.transformer.loras.append(lora)

            if self.clip is not None and output.clip is not None:
                output.clip.loras.append(lora)

            if self.t5_encoder is not None and output.t5_encoder is not None:
                output.t5_encoder.loras.append(lora)

View on GitHub (pinned to 0b6a024f2f)

Solutions

1. Open the workflow in InvokeAI and re-select the LoRA from the model dropdown to refresh its key
2. Re-install/scan the LoRA in the Model Manager so its key exists again
3. Remove the stale LoRA node/field from the graph
4. Check context.models.exists(key) in custom code before invoking

Example fix

// before: hardcoded stale key
lora=ModelIdentifierField(key='old_deleted_lora')
// after: re-pick the model in the UI so the current key is written
lora=ModelIdentifierField(key='<current-key-from-model-manager>')

if not context.models.exists(lora_key):
    raise ValueError(f"LoRA {lora_key} missing from registry; re-select it in the workflow")

def lora_exists(context, lora_key: str) -> bool:
    return context.models.exists(lora_key)

try:
    output = collector.invoke(context)
except Exception as e:
    if str(e).startswith('Unknown lora:'):
        missing_key = str(e).split(':', 1)[1].strip(' !')
        # re-select or remove the LoRA node referencing missing_key
    else:
        raise

When it happens

Trigger: A FLUX LoRA reference field points at a model key that was deleted, purged, or never installed; a stale workflow saved against a since-removed model; a hand-edited graph with an invalid key.

Common situations: Deleting a LoRA in the Model Manager while old workflows still reference it; syncing models across machines; installing InvokeAI fresh and importing old graphs.

Related errors

• Unknown lora: {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.
• LoRA '{lora.lora.key}' is for {lora.lora.base.value if lora.

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/5e5d2103f1b07a93. Report an issue: GitHub.