diff --git a/src/diffusers/modular_pipelines/helios/decoders.py b/src/diffusers/modular_pipelines/helios/decoders.py
index 4f7328c32c7d..c448d36136e6 100644
--- a/src/diffusers/modular_pipelines/helios/decoders.py
+++ b/src/diffusers/modular_pipelines/helios/decoders.py
@@ -76,17 +76,19 @@ def __call__(self, components, state: PipelineState) -> PipelineState:
         block_state = self.get_block_state(state)
 
         vae = components.vae
+        device = components._execution_device
+        decode_dtype = vae.dtype
 
         latents_mean = (
-            torch.tensor(vae.config.latents_mean).view(1, vae.config.z_dim, 1, 1, 1).to(vae.device, vae.dtype)
+            torch.tensor(vae.config.latents_mean).view(1, vae.config.z_dim, 1, 1, 1).to(device, decode_dtype)
         )
         latents_std = 1.0 / torch.tensor(vae.config.latents_std).view(1, vae.config.z_dim, 1, 1, 1).to(
-            vae.device, vae.dtype
+            device, decode_dtype
         )
 
         history_video = None
         for chunk_latents in block_state.latent_chunks:
-            current_latents = chunk_latents.to(vae.dtype) / latents_std + latents_mean
+            current_latents = chunk_latents.to(device=device, dtype=decode_dtype) / latents_std + latents_mean
             current_video = vae.decode(current_latents, return_dict=False)[0]
 
             if history_video is None: