Merge fc2f500571 into c6238047ee

README.md

@@ -183,7 +183,7 @@ Simply download, extract with [7-Zip](https://7-zip.org) or with the windows exp
 
 If you have trouble extracting it, right click the file -> properties -> unblock
 
-Update your Nvidia drivers if it doesn't start.
+The portable above currently comes with python 3.13 and pytorch cuda 13.0. Update your Nvidia drivers if it doesn't start.
 
 #### Alternative Downloads:
 
@@ -212,7 +212,7 @@ Python 3.14 works but you may encounter issues with the torch compile node. The
 
 Python 3.13 is very well supported. If you have trouble with some custom node dependencies on 3.13 you can try 3.12
 
-torch 2.4 and above is supported but some features might only work on newer versions. We generally recommend using the latest major version of pytorch unless it is less than 2 weeks old.
+torch 2.4 and above is supported but some features might only work on newer versions. We generally recommend using the latest major version of pytorch with the latest cuda version unless it is less than 2 weeks old.
 
 ### Instructions:
 

comfy/model_detection.py

@@ -237,6 +237,8 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
         else:
             dit_config["vec_in_dim"] = None
 
+        dit_config["num_heads"] = dit_config["hidden_size"] // sum(dit_config["axes_dim"])
+
         dit_config["depth"] = count_blocks(state_dict_keys, '{}double_blocks.'.format(key_prefix) + '{}.')
         dit_config["depth_single_blocks"] = count_blocks(state_dict_keys, '{}single_blocks.'.format(key_prefix) + '{}.')
         if '{}distilled_guidance_layer.0.norms.0.scale'.format(key_prefix) in state_dict_keys or '{}distilled_guidance_layer.norms.0.scale'.format(key_prefix) in state_dict_keys: #Chroma

comfy/model_management.py

@@ -368,7 +368,7 @@ try:
                     if any((a in arch) for a in ["gfx90a", "gfx942", "gfx1100", "gfx1101", "gfx1151"]):  # TODO: more arches, TODO: gfx950
                         ENABLE_PYTORCH_ATTENTION = True
                 if rocm_version >= (7, 0):
-                   if any((a in arch) for a in ["gfx1201"]):
+                   if any((a in arch) for a in ["gfx1200", "gfx1201"]):
                        ENABLE_PYTORCH_ATTENTION = True
         if torch_version_numeric >= (2, 7) and rocm_version >= (6, 4):
             if any((a in arch) for a in ["gfx1200", "gfx1201", "gfx950"]):  # TODO: more arches, "gfx942" gives error on pytorch nightly 2.10 1013 rocm7.0

comfy/sd.py

@@ -1059,9 +1059,9 @@ def detect_te_model(sd):
         return TEModel.JINA_CLIP_2
     if "encoder.block.23.layer.1.DenseReluDense.wi_1.weight" in sd:
         weight = sd["encoder.block.23.layer.1.DenseReluDense.wi_1.weight"]
-        if weight.shape[-1] == 4096:
+        if weight.shape[0] == 10240:
             return TEModel.T5_XXL
-        elif weight.shape[-1] == 2048:
+        elif weight.shape[0] == 5120:
             return TEModel.T5_XL
     if 'encoder.block.23.layer.1.DenseReluDense.wi.weight' in sd:
         return TEModel.T5_XXL_OLD

comfy/text_encoders/cosmos.py

@@ -36,7 +36,7 @@ def te(dtype_t5=None, t5_quantization_metadata=None):
             if t5_quantization_metadata is not None:
                 model_options = model_options.copy()
                 model_options["t5xxl_quantization_metadata"] = t5_quantization_metadata
-            if dtype is None:
+            if dtype_t5 is not None:
                 dtype = dtype_t5
             super().__init__(device=device, dtype=dtype, model_options=model_options)
     return CosmosTEModel_

comfy/text_encoders/genmo.py

@@ -32,7 +32,7 @@ def mochi_te(dtype_t5=None, t5_quantization_metadata=None):
             if t5_quantization_metadata is not None:
                 model_options = model_options.copy()
                 model_options["t5xxl_quantization_metadata"] = t5_quantization_metadata
-            if dtype is None:
+            if dtype_t5 is not None:
                 dtype = dtype_t5
             super().__init__(device=device, dtype=dtype, model_options=model_options)
     return MochiTEModel_

comfy/text_encoders/pixart_t5.py

@@ -36,7 +36,7 @@ def pixart_te(dtype_t5=None, t5_quantization_metadata=None):
             if t5_quantization_metadata is not None:
                 model_options = model_options.copy()
                 model_options["t5xxl_quantization_metadata"] = t5_quantization_metadata
-            if dtype is None:
+            if dtype_t5 is not None:
                 dtype = dtype_t5
             super().__init__(device=device, dtype=dtype, model_options=model_options)
     return PixArtTEModel_

comfy/utils.py

@@ -30,6 +30,7 @@ from torch.nn.functional import interpolate
 from einops import rearrange
 from comfy.cli_args import args
 import json
+import time
 
 MMAP_TORCH_FILES = args.mmap_torch_files
 DISABLE_MMAP = args.disable_mmap
@@ -1097,6 +1098,10 @@ def set_progress_bar_global_hook(function):
     global PROGRESS_BAR_HOOK
     PROGRESS_BAR_HOOK = function
 
+# Throttle settings for progress bar updates to reduce WebSocket flooding
+PROGRESS_THROTTLE_MIN_INTERVAL = 0.1  # 100ms minimum between updates
+PROGRESS_THROTTLE_MIN_PERCENT = 0.5   # 0.5% minimum progress change
+
 class ProgressBar:
     def __init__(self, total, node_id=None):
         global PROGRESS_BAR_HOOK
@@ -1104,6 +1109,8 @@ class ProgressBar:
         self.current = 0
         self.hook = PROGRESS_BAR_HOOK
         self.node_id = node_id
+        self._last_update_time = 0.0
+        self._last_sent_value = -1
 
     def update_absolute(self, value, total=None, preview=None):
         if total is not None:
@@ -1112,7 +1119,29 @@ class ProgressBar:
             value = self.total
         self.current = value
         if self.hook is not None:
-            self.hook(self.current, self.total, preview, node_id=self.node_id)
+            current_time = time.perf_counter()
+            is_first = (self._last_sent_value < 0)
+            is_final = (value >= self.total)
+            has_preview = (preview is not None)
+
+            # Always send immediately for previews, first update, or final update
+            if has_preview or is_first or is_final:
+                self.hook(self.current, self.total, preview, node_id=self.node_id)
+                self._last_update_time = current_time
+                self._last_sent_value = value
+                return
+
+            # Apply throttling for regular progress updates
+            if self.total > 0:
+                percent_changed = ((value - max(0, self._last_sent_value)) / self.total) * 100
+            else:
+                percent_changed = 100
+            time_elapsed = current_time - self._last_update_time
+
+            if time_elapsed >= PROGRESS_THROTTLE_MIN_INTERVAL and percent_changed >= PROGRESS_THROTTLE_MIN_PERCENT:
+                self.hook(self.current, self.total, preview, node_id=self.node_id)
+                self._last_update_time = current_time
+                self._last_sent_value = value
 
     def update(self, value):
         self.update_absolute(self.current + value)

comfy_api_nodes/nodes_kling.py

@@ -567,7 +567,7 @@ async def execute_lipsync(
     # Upload the audio file to Comfy API and get download URL
     if audio:
         audio_url = await upload_audio_to_comfyapi(
-            cls, audio, container_format="mp3", codec_name="libmp3lame", mime_type="audio/mpeg", filename="output.mp3"
+            cls, audio, container_format="mp3", codec_name="libmp3lame", mime_type="audio/mpeg"
         )
         logging.info("Uploaded audio to Comfy API. URL: %s", audio_url)
     else:

comfy_api_nodes/util/conversions.py

@@ -55,7 +55,7 @@ def image_tensor_pair_to_batch(image1: torch.Tensor, image2: torch.Tensor) -> to
 
 def tensor_to_bytesio(
     image: torch.Tensor,
-    name: str | None = None,
+    *,
     total_pixels: int = 2048 * 2048,
     mime_type: str = "image/png",
 ) -> BytesIO:
@@ -75,7 +75,7 @@ def tensor_to_bytesio(
 
     pil_image = tensor_to_pil(image, total_pixels=total_pixels)
     img_binary = pil_to_bytesio(pil_image, mime_type=mime_type)
-    img_binary.name = f"{name if name else uuid.uuid4()}.{mimetype_to_extension(mime_type)}"
+    img_binary.name = f"{uuid.uuid4()}.{mimetype_to_extension(mime_type)}"
     return img_binary
 
 

comfy_api_nodes/util/upload_helpers.py

@@ -82,7 +82,6 @@ async def upload_audio_to_comfyapi(
     container_format: str = "mp4",
     codec_name: str = "aac",
     mime_type: str = "audio/mp4",
-    filename: str = "uploaded_audio.mp4",
 ) -> str:
     """
     Uploads a single audio input to ComfyUI API and returns its download URL.
@@ -92,7 +91,7 @@ async def upload_audio_to_comfyapi(
     waveform: torch.Tensor = audio["waveform"]
     audio_data_np = audio_tensor_to_contiguous_ndarray(waveform)
     audio_bytes_io = audio_ndarray_to_bytesio(audio_data_np, sample_rate, container_format, codec_name)
-    return await upload_file_to_comfyapi(cls, audio_bytes_io, filename, mime_type)
+    return await upload_file_to_comfyapi(cls, audio_bytes_io, f"{uuid.uuid4()}.{container_format}", mime_type)
 
 
 async def upload_video_to_comfyapi(