invoke-ai/InvokeAI · error · TypeError

Expected ModelPatchRaw for LoRA '{lora.lora.key}', got {type

Error message

Expected ModelPatchRaw for LoRA '{lora.lora.key}', got {type(lora_info.model).__name__}. The LoRA model may be corrupted or incompatible.

What it means

LoRAs applied to the Qwen3 text encoder must be loaded as ModelPatchRaw objects. If context.models.load returns a different model type for the LoRA, the file is not a FLUX.2-compatible LoRA patch or is corrupted, and it cannot be applied to the encoder.

Source

Thrown at invokeai/app/invocations/flux2_klein_text_encoder.py:205

        out = out.to(dtype=text_encoder.dtype, device=device)

        batch_size, num_channels, seq_len, hidden_dim = out.shape
        prompt_embeds = out.permute(0, 2, 1, 3).reshape(batch_size, seq_len, num_channels * hidden_dim)

        last_hidden_state = outputs.hidden_states[-1]
        expanded_mask = attention_mask.unsqueeze(-1).expand_as(last_hidden_state).float()
        sum_embeds = (last_hidden_state * expanded_mask).sum(dim=1)
        num_tokens = expanded_mask.sum(dim=1).clamp(min=1)
        pooled_embeds = sum_embeds / num_tokens

        return prompt_embeds, pooled_embeds

    def _lora_iterator(self, context: InvocationContext) -> Iterator[PatchSpec]:
        """Iterate over LoRA models to apply to the Qwen3 text encoder."""
        for lora in self.qwen3_encoder.loras:
            lora_info = context.models.load(lora.lora)
            if not isinstance(lora_info.model, ModelPatchRaw):
                raise TypeError(
                    f"Expected ModelPatchRaw for LoRA '{lora.lora.key}', got {type(lora_info.model).__name__}. "
                    "The LoRA model may be corrupted or incompatible."
                )
            yield (lora_info.model, lora.weight, lora_info.model_in_ram())

View on GitHub (pinned to 0b6a024f2f)

Solutions

Use a FLUX.2/Klein-compatible LoRA converted for InvokeAI
Re-convert/re-import the LoRA so the model manager stores it as ModelPatchRaw
Check the LoRA model record's type/hash points to the intended file
Re-download the LoRA if the file is corrupted

Example fix

// before: SDXL LoRA wired into qwen3_encoder.loras
loras: [ModelIdentifierField(key='sdxl-lora-abc123')]
// after: FLUX.2-compatible LoRA
loras: [ModelIdentifierField(key='flux2-klein-lora-xyz789')]

Defensive patterns

Strategy: type-guard

Validate before calling

info = context.models.load(lora_field)
if not isinstance(info.model, ModelPatchRaw):
    raise TypeError(f'{lora_field.key} is not a FLUX.2 LoRA patch')

Type guard

from invokeai.backend.model_patcher import ModelPatchRaw

def is_lora_patch(info) -> bool:
    return isinstance(info.model, ModelPatchRaw)

Try / catch

try:
    result = klein_encoder.invoke(context)
except TypeError as e:
    if 'ModelPatchRaw' in str(e):
        convert_lora_to_flux2_format(e)
    raise

Prevention

Only add LoRAs tagged for FLUX.2/Klein to encoder LoRA lists
Re-convert foreign-format LoRAs before use
Verify LoRA records after model-manager migrations

When it happens

Trigger: A LoRA listed in self.qwen3_encoder.loras resolves, via context.models.load(lora.lora), to a model that is not ModelPatchRaw in _lora_iterator; e.g. the record points at a checkpoint-format LoRA or an entirely different model file.

Common situations: Using SD/SDXL LoRA files with FLUX.2 Klein; LoRA converted to a format InvokeAI stores differently; stale model-manager records after conversion; corrupted LoRA downloads.

Related errors

AI-assisted analysis of invoke-ai/InvokeAI@0b6a024f2f (2026-08-29). Data as JSON: /api/errors/544db5ca74a9f741. Report an issue: GitHub.