diff --git a/src/h3_blackwell_runtime/sampler.py b/src/h3_blackwell_runtime/sampler.py
index 6a73dfc..ec2537d 100644
--- a/src/h3_blackwell_runtime/sampler.py
+++ b/src/h3_blackwell_runtime/sampler.py
@@ -51,7 +51,7 @@ def _audio_sigma(video_sigma: torch.Tensor) -> torch.Tensor:
 def sample_video_res_multistep(model, packer: H3PromptPacker, text: torch.Tensor, video: torch.Tensor, audio: torch.Tensor, *, steps: int = 12) -> torch.Tensor:
     """Direct H3 beta/RES sampling with Comfy-equivalent joint AV carry semantics."""
     sigmas = beta_sigmas(steps, device=video.device)
-    audio_carried = audio * 4.0
+    audio_carried = audio
     video_history = audio_history = None
     video_history_sigma = audio_history_sigma = None
     total_steps = len(sigmas) - 1