diff --git a/nodes_model_loading.py b/nodes_model_loading.py
index dedd060..68239da 100644
--- a/nodes_model_loading.py
+++ b/nodes_model_loading.py
@@ -1153,7 +1153,7 @@ class WanVideoModelLoader:
         # currently this can be VACE, MTV-Crafter, Lynx or Ovi-audio weights
         if extra_model is not None:
             for _model in extra_model:
-                print("Loading extra model: ", _model["path"])
+                log.info(f"Loading extra model: {_model['path']}")
                 if gguf:
                     if not _model["path"].endswith(".gguf"):
                         raise ValueError("With GGUF main model the extra model must also be GGUF quantized, if the main model already has VACE included, you can disconnect the extra module loader")
diff --git a/nodes_sampler.py b/nodes_sampler.py
index c2203bd..d6afd6b 100644
--- a/nodes_sampler.py
+++ b/nodes_sampler.py
@@ -875,7 +875,9 @@ class WanVideoSampler:
 
         # LongCat-Avatar
         longcat_ref_latent = None
+        longcat_num_ref_latents = longcat_num_cond_latents = 0
         longcat_avatar_options = image_embeds.get("longcat_avatar_options", None)
+
         if longcat_avatar_options is not None:
             longcat_ref_latent = longcat_avatar_options.get("longcat_ref_latent", None)
             if longcat_ref_latent is not None:
@@ -884,8 +886,10 @@ class WanVideoSampler:
                 seq_len = math.ceil((latent.shape[2] * latent.shape[3]) / 4 * latent.shape[1])
                 insert_len = longcat_ref_latent.shape[1]
                 clean_latent_indices = list(range(0, insert_len)) + [i + insert_len for i in clean_latent_indices]
+                longcat_num_ref_latents = longcat_ref_latent.shape[1]
                 latent_video_length += insert_len
-                log.info(f"LongCat clean_latent_indices: {clean_latent_indices}")
+            longcat_num_cond_latents = len(clean_latent_indices)
+            log.info(f"LongCat num_cond_latents: {longcat_num_cond_latents} num_ref_latents: {longcat_num_ref_latents}")
         audio_stride = 2 if transformer.is_longcat else 1
 
         #controlnet
@@ -1565,8 +1569,8 @@ class WanVideoSampler:
                     "ovi_negative_text_embeds": ovi_negative_text_embeds, # Audio latent model negative text embeds for Ovi
                     "flashvsr_LQ_latent": flashvsr_LQ_latent, # FlashVSR LQ latent for upsampling
                     "flashvsr_strength": flashvsr_strength, # FlashVSR strength
-                    "longcat_num_cond_latents": len(clean_latent_indices) if transformer.is_longcat else 0,
-                    "longcat_num_ref_latents": longcat_ref_latent.shape[1] if longcat_ref_latent is not None else 0,
+                    "longcat_num_cond_latents": longcat_num_cond_latents,
+                    "longcat_num_ref_latents": longcat_num_ref_latents,
                     "longcat_avatar_options": longcat_avatar_options, # LongCat avatar attention options
                     "sdancer_input": sdancer_input, # SteadyDancer input
                     "one_to_all_input": one_to_all_data, # One-to-All input
diff --git a/wanvideo/modules/model.py b/wanvideo/modules/model.py
index 19abc84..89f180f 100644
--- a/wanvideo/modules/model.py
+++ b/wanvideo/modules/model.py
@@ -2210,7 +2210,6 @@ class WanModel(torch.nn.Module):
                 torch.tensor([ref_frame_index], dtype=dtype, device=device),
                 torch.arange(0, steps_t - longcat_num_ref_latents, dtype=dtype, device=device)
             ], dim=0)
-            print("grid_t:", grid_t)
             img_ids[:, :, :, 0] = img_ids[:, :, :, 0] + grid_t.reshape(-1, 1, 1)
         else:
             # Standard temporal encoding
@@ -2639,7 +2638,6 @@ class WanModel(torch.nn.Module):
                 self.cached_key == cache_key):
                 freqs = self.cached_freqs
             else:
-                log.info("Generating new RoPE frequencies")
                 freqs = self.rope_encode_comfy(
                     F, H, W,
                     freq_offset=freq_offset,
@@ -2650,6 +2648,7 @@ class WanModel(torch.nn.Module):
                     device=x.device,
                     dtype=x.dtype
                 )
+                log.info("Generated new RoPE frequencies")
 
                 if s2v_ref_latent is not None:
                     freqs_ref = self.rope_encode_comfy(
@@ -2855,8 +2854,6 @@ class WanModel(torch.nn.Module):
             human_num = len(multitalk_audio_embedding)
 
             # LongCat-Avatar specific
-            tqdm.write(f"longcat_num_cond_latents: {longcat_num_cond_latents}, longcat_num_ref_latents: {longcat_num_ref_latents}")
-
             if longcat_num_ref_latents > 0:
                 audio_start_ref = multitalk_audio_embedding[:, [0], :, :] # padding
                 multitalk_audio_embedding = torch.cat([audio_start_ref, multitalk_audio_embedding], dim=1).contiguous()