Use main diffusers 0.30.0

examples/cogvideo_vid2vid_test_example_01.json

@@ -130,7 +130,7 @@
         "Node name for S&R": "CogVideoTextEncode"
       },
       "widgets_values": [
-        "bad quality video, blurry, messy"
+        ""
       ]
     },
     {
@@ -163,7 +163,7 @@
         "Node name for S&R": "DownloadAndLoadCogVideoModel"
       },
       "widgets_values": [
-        "fp16"
+        "bf16"
       ]
     },
     {

examples/example_01.json

@@ -285,7 +285,7 @@
         "Node name for S&R": "DownloadAndLoadCogVideoModel"
       },
       "widgets_values": [
-        "fp16"
+        "bf16"
       ]
     },
     {

nodes.py

@@ -2,6 +2,7 @@ import os
 import torch
 import folder_paths
 import comfy.model_management as mm
+from comfy.utils import ProgressBar
 from diffusers.schedulers import CogVideoXDDIMScheduler, CogVideoXDPMScheduler
 from diffusers.models import AutoencoderKLCogVideoX, CogVideoXTransformer3DModel
 from .pipeline_cogvideox import CogVideoXPipeline
@@ -24,7 +25,7 @@ class DownloadAndLoadCogVideoModel:
                         "fp32",
                         "bf16",
                     ],
-                    {"default": "fp16"},
+                    {"default": "bf16"},
                 ),
             },
         }
@@ -286,14 +287,16 @@ class CogVideoDecode:
         latents = 1 / vae.config.scaling_factor * latents
 
         frames = []
+        pbar = ProgressBar(num_seconds)
         for i in range(num_seconds):
-            # Whether or not to clear fake context parallel cache
-            fake_cp = i + 1 < num_seconds
             start_frame, end_frame = (0, 3) if i == 0 else (2 * i + 1, 2 * i + 3)
-            current_frames = vae.decode(latents[:, :, start_frame:end_frame], fake_cp=fake_cp).sample
+            current_frames = vae.decode(latents[:, :, start_frame:end_frame]).sample
             frames.append(current_frames)
-            mm.soft_empty_cache()
+            
+            pbar.update(1)
+        vae.clear_fake_context_parallel_cache()
         vae.to(offload_device)
+        mm.soft_empty_cache()
 
         frames = torch.cat(frames, dim=2)
         video = pipeline["pipe"].video_processor.postprocess_video(video=frames, output_type="pt")

readme.md

@@ -1,8 +1,6 @@
 # WORK IN PROGRESS
 
-Currently requires diffusers with PR: https://github.com/huggingface/diffusers/pull/9082
-
-This is specified in requirements.txt
+Requires diffusers 0.30.0 (this is specified in requirements.txt)
 
 
 https://github.com/user-attachments/assets/9e41f37b-2bb3-411c-81fa-e91b80da2559

requirements.txt

@@ -1,2 +1,2 @@
 huggingface_hub
-git+https://github.com/huggingface/diffusers.git@878f609aa5ce4a78fea0f048726889debde1d7e8#egg=diffusers
+diffusers>=0.30.0