feat(lora-loader): Load LoRAs using lower-level API to bypass folder_paths validation

- Add get_lora_info_absolute() function to return absolute file paths
- Replace LoraLoader().load_lora() with comfy.utils.load_torch_file() +
  comfy.sd.load_lora_for_models() to enable loading LoRAs from any path
- This allows LoRA Manager to load LoRAs from non-standard paths (multi-library support)
- Fixes #805

py/nodes/lora_loader.py

@@ -1,7 +1,8 @@
 import logging
 import re
-from nodes import LoraLoader
-from ..utils.utils import get_lora_info
+import comfy.utils # type: ignore
+import comfy.sd # type: ignore
+from ..utils.utils import get_lora_info_absolute
 from .utils import FlexibleOptionalInputType, any_type, extract_lora_name, get_loras_list, nunchaku_load_lora
 
 logger = logging.getLogger(__name__)
@@ -58,12 +59,13 @@ class LoraLoaderLM:
                     model = nunchaku_load_lora(model, lora_path, model_strength)
                     # clip remains unchanged for Nunchaku models
                 else:
-                    # Use default loader for standard models
-                    model, clip = LoraLoader().load_lora(model, clip, lora_path, model_strength, clip_strength)
+                    # Use lower-level API to load LoRA directly without folder_paths validation
+                    lora = comfy.utils.load_torch_file(lora_path, safe_load=True)
+                    model, clip = comfy.sd.load_lora_for_models(model, clip, lora, model_strength, clip_strength)
                 
                 # Extract lora name for trigger words lookup
                 lora_name = extract_lora_name(lora_path)
-                _, trigger_words = get_lora_info(lora_name)
+                _, trigger_words = get_lora_info_absolute(lora_name)
                 
                 all_trigger_words.extend(trigger_words)
                 # Add clip strength to output if different from model strength (except for Nunchaku models)
@@ -84,7 +86,7 @@ class LoraLoaderLM:
             clip_strength = float(lora.get('clipStrength', model_strength))
             
             # Get lora path and trigger words
-            lora_path, trigger_words = get_lora_info(lora_name)
+            lora_path, trigger_words = get_lora_info_absolute(lora_name)
             
             # Apply the LoRA using the appropriate loader
             if is_nunchaku_model:
@@ -92,8 +94,9 @@ class LoraLoaderLM:
                 model = nunchaku_load_lora(model, lora_path, model_strength)
                 # clip remains unchanged
             else:
-                # Use default loader for standard models
-                model, clip = LoraLoader().load_lora(model, clip, lora_path, model_strength, clip_strength)
+                # Use lower-level API to load LoRA directly without folder_paths validation
+                lora = comfy.utils.load_torch_file(lora_path, safe_load=True)
+                model, clip = comfy.sd.load_lora_for_models(model, clip, lora, model_strength, clip_strength)
             
             # Include clip strength in output if different from model strength and not a Nunchaku model
             if not is_nunchaku_model and abs(model_strength - clip_strength) > 0.001:
@@ -199,12 +202,13 @@ class LoraTextLoaderLM:
                     model = nunchaku_load_lora(model, lora_path, model_strength)
                     # clip remains unchanged for Nunchaku models
                 else:
-                    # Use default loader for standard models
-                    model, clip = LoraLoader().load_lora(model, clip, lora_path, model_strength, clip_strength)
+                    # Use lower-level API to load LoRA directly without folder_paths validation
+                    lora = comfy.utils.load_torch_file(lora_path, safe_load=True)
+                    model, clip = comfy.sd.load_lora_for_models(model, clip, lora, model_strength, clip_strength)
                 
                 # Extract lora name for trigger words lookup
                 lora_name = extract_lora_name(lora_path)
-                _, trigger_words = get_lora_info(lora_name)
+                _, trigger_words = get_lora_info_absolute(lora_name)
                 
                 all_trigger_words.extend(trigger_words)
                 # Add clip strength to output if different from model strength (except for Nunchaku models)
@@ -221,7 +225,7 @@ class LoraTextLoaderLM:
             clip_strength = lora['clip_strength']
             
             # Get lora path and trigger words
-            lora_path, trigger_words = get_lora_info(lora_name)
+            lora_path, trigger_words = get_lora_info_absolute(lora_name)
             
             # Apply the LoRA using the appropriate loader
             if is_nunchaku_model:
@@ -229,8 +233,9 @@ class LoraTextLoaderLM:
                 model = nunchaku_load_lora(model, lora_path, model_strength)
                 # clip remains unchanged
             else:
-                # Use default loader for standard models
-                model, clip = LoraLoader().load_lora(model, clip, lora_path, model_strength, clip_strength)
+                # Use lower-level API to load LoRA directly without folder_paths validation
+                lora = comfy.utils.load_torch_file(lora_path, safe_load=True)
+                model, clip = comfy.sd.load_lora_for_models(model, clip, lora, model_strength, clip_strength)
             
             # Include clip strength in output if different from model strength and not a Nunchaku model
             if not is_nunchaku_model and abs(model_strength - clip_strength) > 0.001:

py/utils/utils.py

@@ -50,6 +50,52 @@ def get_lora_info(lora_name):
         # No event loop is running, we can use asyncio.run()
         return asyncio.run(_get_lora_info_async())
 
+
+def get_lora_info_absolute(lora_name):
+    """Get the absolute lora path and trigger words from cache
+    
+    Returns:
+        tuple: (absolute_path, trigger_words) where absolute_path is the full 
+               file system path to the LoRA file, or original lora_name if not found
+    """
+    async def _get_lora_info_absolute_async():
+        scanner = await ServiceRegistry.get_lora_scanner()
+        cache = await scanner.get_cached_data()
+        
+        for item in cache.raw_data:
+            if item.get('file_name') == lora_name:
+                file_path = item.get('file_path')
+                if file_path:
+                    # Return absolute path directly
+                    # Get trigger words from civitai metadata
+                    civitai = item.get('civitai', {})
+                    trigger_words = civitai.get('trainedWords', []) if civitai else []
+                    return file_path, trigger_words
+        return lora_name, []
+    
+    try:
+        # Check if we're already in an event loop
+        loop = asyncio.get_running_loop()
+        # If we're in a running loop, we need to use a different approach
+        # Create a new thread to run the async code
+        import concurrent.futures
+        
+        def run_in_thread():
+            new_loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(new_loop)
+            try:
+                return new_loop.run_until_complete(_get_lora_info_absolute_async())
+            finally:
+                new_loop.close()
+        
+        with concurrent.futures.ThreadPoolExecutor() as executor:
+            future = executor.submit(run_in_thread)
+            return future.result()
+            
+    except RuntimeError:
+        # No event loop is running, we can use asyncio.run()
+        return asyncio.run(_get_lora_info_absolute_async())
+
 def fuzzy_match(text: str, pattern: str, threshold: float = 0.85) -> bool:
         """
         Check if text matches pattern using fuzzy matching.