diff --git a/cosmos_predict2/_src/predict2/action/inference/inference.py b/cosmos_predict2/_src/predict2/action/inference/inference.py
index e30929f91..8e373699f 100644
--- a/cosmos_predict2/_src/predict2/action/inference/inference.py
+++ b/cosmos_predict2/_src/predict2/action/inference/inference.py
@@ -360,7 +360,9 @@ def main():
             if args.single_chunk:
                 break
 
-        chunk_list = [chunk_video[0]] + [chunk_video[i][: args.chunk_size] for i in range(1, len(chunk_video))]
+        # Drop the first frame of each subsequent chunk: it is the reconstructed
+        # conditioning frame (~ previous chunk's last frame) and would otherwise duplicate.
+        chunk_list = [chunk_video[0]] + [chunk_video[i][1:] for i in range(1, len(chunk_video))]
         chunk_video = np.concatenate(chunk_list, axis=0)
         if args.single_chunk:
             chunk_video_name = f"{args.save_root}/{img_name + '_single_chunk.mp4'}"
diff --git a/cosmos_predict2/_src/predict2/action/inference/inference_gr00t.py b/cosmos_predict2/_src/predict2/action/inference/inference_gr00t.py
index 7efc2be06..881308a3b 100644
--- a/cosmos_predict2/_src/predict2/action/inference/inference_gr00t.py
+++ b/cosmos_predict2/_src/predict2/action/inference/inference_gr00t.py
@@ -337,7 +337,9 @@ def main():
             if args.single_chunk:
                 break
 
-        chunk_list = [chunk_video[0]] + [chunk_video[i][: args.chunk_size] for i in range(1, len(chunk_video))]
+        # Drop the first frame of each subsequent chunk: it is the reconstructed
+        # conditioning frame (~ previous chunk's last frame) and would otherwise duplicate.
+        chunk_list = [chunk_video[0]] + [chunk_video[i][1:] for i in range(1, len(chunk_video))]
         chunk_video = np.concatenate(chunk_list, axis=0)
         if args.single_chunk:
             chunk_video_name = f"{args.save_root}/{img_name + '_single_chunk.mp4'}"
diff --git a/cosmos_predict2/action_conditioned.py b/cosmos_predict2/action_conditioned.py
index 45dcfc4b0..727557382 100644
--- a/cosmos_predict2/action_conditioned.py
+++ b/cosmos_predict2/action_conditioned.py
@@ -358,9 +358,9 @@ def inference(
             if inference_args.single_chunk:
                 break
 
-        chunk_list = [chunk_video[0]] + [
-            chunk_video[i][: inference_args.chunk_size] for i in range(1, len(chunk_video))
-        ]
+        # Drop the first frame of each subsequent chunk: it is the reconstructed
+        # conditioning frame (~ previous chunk's last frame) and would otherwise duplicate.
+        chunk_list = [chunk_video[0]] + [chunk_video[i][1:] for i in range(1, len(chunk_video))]
         chunk_video = np.concatenate(chunk_list, axis=0)
         if inference_args.single_chunk:
             chunk_video_name = str(inference_args.save_root / f"{img_name}_single_chunk.mp4")