some fixes in model loading and nodes

comfy/ldm/hunyuan_foley/vae.py

@@ -5,7 +5,7 @@ from typing import List
 import torch.nn as nn
 from einops import rearrange
 from torchvision.transforms import v2
-from torch.nn.utils import weight_norm
+from torch.nn.utils.parametrizations import weight_norm
 
 from comfy.ldm.hunyuan_foley.syncformer import Synchformer
 
@@ -96,12 +96,6 @@ class DACEncoder(nn.Module):
             d_model *= 2
             self.block += [DACEncoderBlock(d_model, stride=stride, device = device, dtype = dtype, operations = operations)]
 
-        # Create last convolution
-        self.block += [
-            Snake1d(d_model),
-            WNConv1d(d_model, d_latent, kernel_size=3, padding=1, device = device, dtype = dtype, operations = operations),
-        ]
-
         # Wrap black into nn.Sequential
         self.block = nn.Sequential(*self.block)
         self.enc_dim = d_model
@@ -151,12 +145,6 @@ class DACDecoder(nn.Module):
             output_dim = channels // 2 ** (i + 1)
             layers += [DACDecoderBlock(input_dim, output_dim, stride, device = device, dtype = dtype, operations = operations)]
 
-        layers += [
-            Snake1d(output_dim, device = device, dtype = dtype),
-            WNConv1d(output_dim, d_out, kernel_size=7, padding=3, device = device, dtype = dtype, operations = operations),
-            nn.Tanh(),
-        ]
-
         self.model = nn.Sequential(*layers)
 
     def forward(self, x):

comfy/supported_models.py

@@ -1319,6 +1319,10 @@ class HunyuanFoley(supported_models_base.BASE):
     def clip_target(self, state_dict={}):
         return supported_models_base.ClipTarget(comfy.text_encoders.clap_model.ClapLargeTokenizer, comfy.text_encoders.clap_model.ClapTextEncoderModel)
     
+    def process_clip_state_dict(self, state_dict):
+        state_dict = utils.state_dict_prefix_replace(state_dict, {k: "transformer." for k in self.text_encoder_key_prefix}, filter_keys=True)
+        return state_dict
+
 class QwenImage(supported_models_base.BASE):
     unet_config = {
         "image_model": "qwen_image",

comfy_extras/nodes_hunyuan_foley.py

@@ -34,8 +34,8 @@ class HunyuanFoleyConditioning(io.ComfyNode):
             display_name="HunyuanFoleyConditioning",
             category="conditioning/video_models",
             inputs = [
-                io.Conditioning.Input("video_encoding_1"),
+                io.Conditioning.Input("siglip_encoding_1"),
-                io.Conditioning.Input("video_encoding_2"),
+                io.Conditioning.Input("synchformer_encoding_2"),
                 io.Conditioning.Input("text_encoding"),
             ],
             outputs=[io.Conditioning.Output(display_name= "positive"), io.Conditioning.Output(display_name="negative")]

comfy_extras/nodes_video.py

@@ -14,6 +14,7 @@ from comfy_api.input_impl import VideoFromComponents, VideoFromFile
 from comfy_api.util import VideoCodec, VideoComponents, VideoContainer
 from comfy_api.latest import ComfyExtension, io, ui
 from comfy.cli_args import args
+import comfy.utils
 
 class EncodeVideo(io.ComfyNode):
     @classmethod
@@ -49,6 +50,7 @@ class EncodeVideo(io.ComfyNode):
     
     @classmethod
     def execute(cls, video, processing_batch_size, step_size, vae = None, clip_vision = None):
+
         t, c, h, w = video.shape
         b = 1
         batch_size = b * t
@@ -71,10 +73,15 @@ class EncodeVideo(io.ComfyNode):
 
         outputs = []
         total = data.shape[0]
+        pbar = comfy.utils.ProgressBar(total/batch_size)
+        with torch.inference_mode(): 
             for i in range(0, total, batch_size):
                 chunk = data[i : i + batch_size]
                 out = vae.encode(chunk)
                 outputs.append(out)
+                del out, chunk
+                torch.cuda.empty_cache()
+                pbar.update(1)
 
         output = torch.cat(outputs)
 
@@ -109,7 +116,7 @@ class ResampleVideo(io.ComfyNode):
                     for frame in packet.decode():
                         arr = torch.from_numpy(frame.to_ndarray(format="rgb24")).float() / 255.0
                         frames.append(arr)
-                return torch.stack(frames)
+                return io.NodeOutput(torch.stack(frames))
 
             stream.thread_type = "AUTO"