Fix base model detection for Wan LoRA adapters

For Wan LoRA adapters, skip reading base_model from stored config
(which may have incorrect values like T2V instead of I2V).
Instead, always infer from LoRA ID by checking for 'i2v' in the name.

This ensures wan2_2_i2v_general_nsfw_lora uses the correct
Wan-AI/Wan2.2-I2V-A14B-Diffusers base model instead of the
incorrect Wan-AI/Wan2.2-T2V-A14B-Diffusers.

videogen.py

@@ -8654,7 +8654,14 @@ def main(args):
     
     if is_lora:
         lora_id = m_info["id"]
-        base_model_id = m_info.get("base_model")
+        # For Wan LoRA adapters, don't use stored base_model as it may be incorrect
+        # Instead, always infer from LoRA ID (check for i2v vs t2v)
+        lora_id_lower = lora_id.lower()
+        if "wan" in lora_id_lower:
+            # Skip reading from config - infer from LoRA ID instead
+            base_model_id = None
+        else:
+            base_model_id = m_info.get("base_model")
         
         # Allow manual override via --base-model
         if args.base_model: