diff --git a/studio/backend/core/inference/inference.py b/studio/backend/core/inference/inference.py
index e90e6c0c2a..1147c281b7 100644
--- a/studio/backend/core/inference/inference.py
+++ b/studio/backend/core/inference/inference.py
@@ -6,8 +6,10 @@ from unsloth.chat_templates import get_chat_template
 from transformers import TextStreamer
 from peft import PeftModel, PeftModelForCausalLM
 
+import json
 import sys
 import torch
+from pathlib import Path
 from typing import Optional, Union, Generator, Tuple
 from utils.models import ModelConfig, get_base_model_from_lora
 from utils.paths import is_model_cached
@@ -112,7 +114,18 @@ class InferenceBackend:
                 # In that case, load the real processor from the base model.
                 from transformers import ProcessorMixin
                 if not (isinstance(processor, ProcessorMixin) or hasattr(processor, "image_processor")):
+                    # For LoRA adapters, use the base model. For local merged exports,
+                    # read export_metadata.json to find the original base model.
                     processor_source = config.base_model if config.is_lora else config.identifier
+                    if not config.is_lora and config.is_local:
+                        _meta_path = Path(config.path) / "export_metadata.json"
+                        try:
+                            if _meta_path.exists():
+                                _meta = json.loads(_meta_path.read_text())
+                                if _meta.get("base_model"):
+                                    processor_source = _meta["base_model"]
+                        except Exception:
+                            pass
                     logger.warning(
                         f"FastVisionModel returned {type(processor).__name__} (no image_processor) "
                         f"for '{model_name}' — loading proper processor from '{processor_source}'"