diff --git a/modules/sd_samplers_common.py b/modules/sd_samplers_common.py
index c060cccb2..b475b9f81 100644
--- a/modules/sd_samplers_common.py
+++ b/modules/sd_samplers_common.py
@@ -1,6 +1,6 @@
 import inspect
 from collections import namedtuple
-import numpy as np
+from contextlib import nullcontext
 import torch
 from PIL import Image
 from modules import devices, images, sd_vae_approx, sd_samplers, sd_vae_taesd, shared, sd_models
@@ -59,15 +59,11 @@ def samples_to_images_tensor(sample, approximation=None, model=None):
 
     return x_sample
 
-
-def single_sample_to_image(sample, approximation=None):
+def single_sample_to_image(sample, approximation=None, non_blocking=False):
     x_sample = samples_to_images_tensor(sample.unsqueeze(0), approximation)[0] * 0.5 + 0.5
-
     x_sample = torch.clamp(x_sample, min=0.0, max=1.0)
-    x_sample = 255. * np.moveaxis(x_sample.cpu().numpy(), 0, 2)
-    x_sample = x_sample.astype(np.uint8)
-
-    return Image.fromarray(x_sample)
+    x_sample = 255. * x_sample.permute(1, 2, 0)
+    return x_sample.to(device='cpu', dtype=torch.uint8, non_blocking=non_blocking)
 
 
 def decode_first_stage(model, x):
@@ -76,12 +72,27 @@ def decode_first_stage(model, x):
     return samples_to_images_tensor(x, approx_index, model)
 
 
+if torch.cuda.is_available():
+    lp_stream = torch.cuda.Stream()
+    live_preview_stream_context = torch.cuda.stream(lp_stream)
+else:
+    lp_stream = None
+    live_preview_stream_context = nullcontext()
+
 def sample_to_image(samples, index=0, approximation=None):
-    return single_sample_to_image(samples[index], approximation)
+    with live_preview_stream_context:
+        sample = single_sample_to_image(samples[index], approximation, non_blocking=lp_stream is not None)
+    if lp_stream is not None:
+        lp_stream.synchronize()
+    return Image.fromarray(sample.numpy())
 
 
 def samples_to_image_grid(samples, approximation=None):
-    return images.image_grid([single_sample_to_image(sample, approximation) for sample in samples])
+    with live_preview_stream_context:
+        sample_tensors = [single_sample_to_image(sample, approximation, non_blocking=lp_stream is not None) for sample in samples]
+    if lp_stream is not None:
+        lp_stream.synchronize()
+    return images.image_grid([Image.fromarray(sample.numpy()) for sample in sample_tensors])
 
 
 def images_tensor_to_samples(image, approximation=None, model=None):