fix: load AutoProcessor from tokenizer subdirectory, not root model path

Co-authored-by: lstein <111189+lstein@users.noreply.github.com>
Agent-Logs-Url: https://github.com/lstein/InvokeAI/sessions/174390c9-a290-4a1b-83cc-c55441b2d9b8

Author: Copilot

invokeai/app/invocations/qwen_image_edit_text_encoder.py

@@ -74,8 +74,10 @@ def _encode(self, context: InvocationContext, images: list[PILImage.Image]) -> t
         # Load the full processor (image_processor + tokenizer) from the tokenizer submodel path.
         # Using AutoProcessor.from_pretrained ensures all components are loaded correctly
         # regardless of whether the model uses Qwen2VLProcessor or Qwen2_5_VLProcessor.
+        # For diffusers models the processor config lives in the `tokenizer` subdirectory,
+        # so we append the submodel directory name to the root model path.
         tokenizer_config = context.models.get_config(self.qwen_vl_encoder.tokenizer)
-        tokenizer_abs_path = context.models.get_absolute_path(tokenizer_config)
+        tokenizer_abs_path = context.models.get_absolute_path(tokenizer_config) / "tokenizer"
         processor = AutoProcessor.from_pretrained(str(tokenizer_abs_path), local_files_only=True)
 
         text_encoder_info = context.models.load(self.qwen_vl_encoder.text_encoder)