Fix: Wrap LoRA loading and offloading in defer_i2v_loading check

When defer_i2v_loading=True (I2V mode without provided image), the code
sets pipe=None but then tried to call pipe.load_lora_weights() and
pipe.enable_model_cpu_offload() on None, causing AttributeError.

This fix wraps the LoRA loading and offloading configuration blocks
inside an 'if not defer_i2v_loading:' condition so they are skipped
when the I2V model loading is deferred until after image generation.

videogen

@@ -3938,61 +3938,63 @@ def main(args):
     
     timing.end_step()  # model_loading
     
-    # Apply LoRA if this is a LoRA model
-    if is_lora and lora_id:
-        timing.begin_step("lora_loading")
-        print(f"  Loading LoRA adapter: {lora_id}")
-        try:
-            # Load LoRA weights
-            pipe.load_lora_weights(lora_id)
-            print(f"  ✅ LoRA applied successfully")
-        except Exception as e:
-            print(f"  ⚠️ LoRA loading failed: {e}")
-            print(f"     Continuing with base model...")
-        timing.end_step()  # lora_loading
+    # Only apply LoRA and offloading if we actually loaded the model (not deferred)
+    if not defer_i2v_loading:
+        # Apply LoRA if this is a LoRA model
+        if is_lora and lora_id:
+            timing.begin_step("lora_loading")
+            print(f"  Loading LoRA adapter: {lora_id}")
+            try:
+                # Load LoRA weights
+                pipe.load_lora_weights(lora_id)
+                print(f"  ✅ LoRA applied successfully")
+            except Exception as e:
+                print(f"  ⚠️ LoRA loading failed: {e}")
+                print(f"     Continuing with base model...")
+            timing.end_step()  # lora_loading
 
-    if args.no_filter and hasattr(pipe, "safety_checker"):
-        pipe.safety_checker = None
+        if args.no_filter and hasattr(pipe, "safety_checker"):
+            pipe.safety_checker = None
 
-    # Offloading
-    off = args.offload_strategy
-    if off == "auto_map":
-        pipe.enable_model_cpu_offload()
-    elif off == "sequential":
-        pipe.enable_sequential_cpu_offload()
-    elif off == "group":
-        try:
-            pipe.enable_group_offload(group_size=args.offload_group_size)
-        except:
-            print("Group offload unavailable → model offload fallback")
+        # Offloading
+        off = args.offload_strategy
+        if off == "auto_map":
             pipe.enable_model_cpu_offload()
-    elif off == "model":
-        pipe.enable_model_cpu_offload()
-    else:
-        pipe.to("cuda" if torch.cuda.is_available() else "cpu")
-
-    pipe.enable_attention_slicing("max")
-    try:
-        pipe.enable_vae_slicing()
-        pipe.enable_vae_tiling()
-    except:
-        pass
+        elif off == "sequential":
+            pipe.enable_sequential_cpu_offload()
+        elif off == "group":
+            try:
+                pipe.enable_group_offload(group_size=args.offload_group_size)
+            except:
+                print("Group offload unavailable → model offload fallback")
+                pipe.enable_model_cpu_offload()
+        elif off == "model":
+            pipe.enable_model_cpu_offload()
+        else:
+            pipe.to("cuda" if torch.cuda.is_available() else "cpu")
 
-    if torch.cuda.is_available():
+        pipe.enable_attention_slicing("max")
         try:
-            pipe.enable_xformers_memory_efficient_attention()
+            pipe.enable_vae_slicing()
+            pipe.enable_vae_tiling()
         except:
             pass
 
-    if "wan" in args.model and hasattr(pipe, "scheduler"):
-        try:
-            pipe.scheduler = UniPCMultistepScheduler.from_config(
-                pipe.scheduler.config,
-                prediction_type="flow_prediction",
-                flow_shift=extra.get("flow_shift", 3.0)
-            )
-        except:
-            pass
+        if torch.cuda.is_available():
+            try:
+                pipe.enable_xformers_memory_efficient_attention()
+            except:
+                pass
+
+        if "wan" in args.model and hasattr(pipe, "scheduler"):
+            try:
+                pipe.scheduler = UniPCMultistepScheduler.from_config(
+                    pipe.scheduler.config,
+                    prediction_type="flow_prediction",
+                    flow_shift=extra.get("flow_shift", 3.0)
+                )
+            except:
+                pass
 
     # ─── Generation ────────────────────────────────────────────────────────────
     seed = args.seed if args.seed >= 0 else random.randint(0, 2**31 - 1)