Merge branch 'master' into v3-improvements

comfy/ldm/lumina/model.py

@@ -491,7 +491,8 @@ class NextDiT(nn.Module):
                 for layer_id in range(n_layers)
             ]
         )
-        self.norm_final = operation_settings.get("operations").RMSNorm(dim, eps=norm_eps, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
+        # This norm final is in the lumina 2.0 code but isn't actually used for anything.
+        # self.norm_final = operation_settings.get("operations").RMSNorm(dim, eps=norm_eps, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
         self.final_layer = FinalLayer(dim, patch_size, self.out_channels, z_image_modulation=z_image_modulation, operation_settings=operation_settings)
 
         if self.pad_tokens_multiple is not None:

comfy/model_management.py

@@ -1019,8 +1019,8 @@ NUM_STREAMS = 0
 if args.async_offload is not None:
     NUM_STREAMS = args.async_offload
 else:
-    #  Enable by default on Nvidia
+    #  Enable by default on Nvidia and AMD
-    if is_nvidia():
+    if is_nvidia() or is_amd():
         NUM_STREAMS = 2
 
 if args.disable_async_offload:
@@ -1126,6 +1126,16 @@ if not args.disable_pinned_memory:
 
 PINNING_ALLOWED_TYPES = set(["Parameter", "QuantizedTensor"])
 
+def discard_cuda_async_error():
+    try:
+        a = torch.tensor([1], dtype=torch.uint8, device=get_torch_device())
+        b = torch.tensor([1], dtype=torch.uint8, device=get_torch_device())
+        _ = a + b
+        torch.cuda.synchronize()
+    except torch.AcceleratorError:
+        #Dump it! We already know about it from the synchronous return
+        pass
+
 def pin_memory(tensor):
     global TOTAL_PINNED_MEMORY
     if MAX_PINNED_MEMORY <= 0:
@@ -1158,6 +1168,9 @@ def pin_memory(tensor):
         PINNED_MEMORY[ptr] = size
         TOTAL_PINNED_MEMORY += size
         return True
+    else:
+        logging.warning("Pin error.")
+        discard_cuda_async_error()
 
     return False
 
@@ -1186,6 +1199,9 @@ def unpin_memory(tensor):
         if len(PINNED_MEMORY) == 0:
             TOTAL_PINNED_MEMORY = 0
         return True
+    else:
+        logging.warning("Unpin error.")
+        discard_cuda_async_error()
 
     return False
 

comfy_api_nodes/nodes_veo2.py

@@ -168,6 +168,8 @@ class VeoVideoGenerationNode(IO.ComfyNode):
         # Only add generateAudio for Veo 3 models
         if model.find("veo-2.0") == -1:
             parameters["generateAudio"] = generate_audio
+            # force "enhance_prompt" to True for Veo3 models
+            parameters["enhancePrompt"] = True
 
         initial_response = await sync_op(
             cls,
@@ -291,7 +293,7 @@ class Veo3VideoGenerationNode(VeoVideoGenerationNode):
                 IO.Boolean.Input(
                     "enhance_prompt",
                     default=True,
-                    tooltip="Whether to enhance the prompt with AI assistance",
+                    tooltip="This parameter is deprecated and ignored.",
                     optional=True,
                 ),
                 IO.Combo.Input(

comfy_api_nodes/util/client.py

@@ -430,9 +430,9 @@ def _display_text(
     if status:
         display_lines.append(f"Status: {status.capitalize() if isinstance(status, str) else status}")
     if price is not None:
-        p = f"{float(price):,.4f}".rstrip("0").rstrip(".")
+        p = f"{float(price) * 211:,.1f}".rstrip("0").rstrip(".")
         if p != "0":
-            display_lines.append(f"Price: ${p}")
+            display_lines.append(f"Price: {p} credits")
     if text is not None:
         display_lines.append(text)
     if display_lines: