All checks were successful
Build and Push RunPod Docker Image / build-and-push (push) Successful in 13s
The correct function to patch is decode_audio from infer_utils module, which is where chunked VAE decoding actually happens. This intercepts the call at the right level to force chunked=False. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
48 lines
1.3 KiB
Python
48 lines
1.3 KiB
Python
"""
|
|
Pivoine DiffRhythm Node
|
|
Custom wrapper for DiffRhythm that disables chunked decoding to prevent
|
|
tensor dimension mismatch errors (32 vs 64) in VAE overlap logic.
|
|
|
|
Author: valknar@pivoine.art
|
|
"""
|
|
|
|
import sys
|
|
sys.path.append('/workspace/ComfyUI/custom_nodes/ComfyUI_DiffRhythm')
|
|
|
|
# Monkey-patch decode_audio from infer_utils to force chunked=False
|
|
import infer_utils
|
|
_original_decode_audio = infer_utils.decode_audio
|
|
|
|
def patched_decode_audio(latent, vae_model, chunked=True):
|
|
"""Patched version that always uses chunked=False"""
|
|
return _original_decode_audio(latent, vae_model, chunked=False)
|
|
|
|
# Apply the monkey patch
|
|
infer_utils.decode_audio = patched_decode_audio
|
|
|
|
from DiffRhythmNode import DiffRhythmRun
|
|
|
|
class PivoineDiffRhythmRun(DiffRhythmRun):
|
|
"""
|
|
Pivoine version of DiffRhythmRun with chunked decoding forcibly disabled.
|
|
|
|
Changes from original:
|
|
- Monkey-patches the infer() function to always use chunked=False
|
|
- Prevents tensor dimension mismatch in VAE (32 vs 64 channel error)
|
|
- Requires more VRAM (~12-16GB) but works reliably on RTX 4090
|
|
"""
|
|
|
|
CATEGORY = "🌸Pivoine/Audio"
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return super().INPUT_TYPES()
|
|
|
|
NODE_CLASS_MAPPINGS = {
|
|
"PivoineDiffRhythmRun": PivoineDiffRhythmRun,
|
|
}
|
|
|
|
NODE_DISPLAY_NAME_MAPPINGS = {
|
|
"PivoineDiffRhythmRun": "Pivoine DiffRhythm Run",
|
|
}
|