Restore colorspace detection using loaded model

Detection now uses the same model instance that's already in memory
with offload strategy configured. The detection respects the model's
memory management setup.

videogen.py

@@ -1354,8 +1354,11 @@ def detect_model_colorspace(pipe, model_name, m_info, args):
 def get_model_colorspace(pipe, model_name, m_info, args):
     """Get the colorspace for a model, detecting it if necessary.
     
+    Uses the already-loaded model to generate a minimal test frame.
+    The detection respects the model's offload strategy.
+    
     Args:
-        pipe: The loaded diffusion pipeline
+        pipe: The loaded diffusion pipeline (already configured with offload)
         model_name: Name of the model in MODELS config
         m_info: Model info dict from MODELS
         args: Command line arguments
@@ -1367,29 +1370,8 @@ def get_model_colorspace(pipe, model_name, m_info, args):
     if model_name in MODELS and "colorspace" in MODELS[model_name]:
         return MODELS[model_name]["colorspace"]
     
-    # Skip detection if using offload strategies (memory-constrained)
-    # Colorspace detection requires extra memory that may not be available
-    offload_strategy = getattr(args, 'offload_strategy', 'model')
-    if offload_strategy not in ['none', 'model']:
-        print(f"  📊 Colorspace: Skipping detection (offload_strategy={offload_strategy})")
-        print(f"     Defaulting to RGB (use --swap_bgr if colors look wrong)")
-        # Save RGB as default to avoid repeated attempts
-        if model_name in MODELS:
-            MODELS[model_name]["colorspace"] = "RGB"
-            save_models_config(MODELS)
-        return "RGB"
-    
-    # Also skip if low system RAM
-    available_ram_gb = get_available_ram_gb()
-    if available_ram_gb < 12.0:  # Need ample RAM for detection
-        print(f"  📊 Colorspace: Skipping detection (low RAM: {available_ram_gb:.1f}GB)")
-        print(f"     Defaulting to RGB (use --swap_bgr if colors look wrong)")
-        if model_name in MODELS:
-            MODELS[model_name]["colorspace"] = "RGB"
-            save_models_config(MODELS)
-        return "RGB"
-    
-    # Detect and save
+    # Run detection using the same model that's already loaded
+    # This respects the offload strategy already configured
     return detect_model_colorspace(pipe, model_name, m_info, args)