Hunyuan Video Foley ComfyUI Workflow: Turn Quiet Video into Sound

Esha
By Esha
3 Min Read

i found a model called hunyuan video foley. you give it a quiet video, it makes the sound that fits: wind that pushes, fire that crackles, rocks that rumble, even a light music bed when the scene needs it.

I tried a few shots and it surprised me. i dropped in a beach clip with no audio, hit run

how I set it up in comfyui, then a few quick runs so you can hear what it does.

there isn’t an official comfyui node yet. i made a small custom one so we can test. repo: https://github.com/aistudynow/Comfyui-HunyuanFoley
the weights live on the model page: https://huggingface.co/tencent/HunyuanVideo-Foley/tree/main

Setup I Used

open ComfyUI/custom_nodes/ in a terminal, run

git clone https://github.com/aistudynow/Comfyui-HunyuanFoley.git

wait for it to finish. go into the new Custom node/HunyuanVideo-Foley folder. Open the Folder in cmd

pip install -r requirements.txt

Also

Open a terminal in G:\ComfyUI_windows_portable\:

.\python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI_HunyuanFoley\requirements.txt

inside the node, make a folder models/hunyuanvideo_foley/ if you don’t see it. put these four files there and keep names exact: config.yaml, hunyuanvideo_foley.pth, synchformer_state_dict.pth, vae_128d_48k.pth.

back in comfyui, set the paths in the node:
main model → hunyuanvideo_foley.pth
vae → vae_128d_48k.pth
syncformer → synchformer_state_dict.pth
config → config.yaml

if a path is wrong you’ll get a red warning. point it to the right file and you’re fine.

The Small Graph

load the video with Load Video. a short hint is enough: “ocean sound”, “burning tree”, “windy street”. match frame rate in two spots: set fps in the hunyuanvideo-foley sampler, and set the Combine Video node to the same number. so 24/24, or 30/30, etc. when it looks right, press Generate.

if your graph also rebuilds the picture and you see color flicker, leave any color-match step off on that pass. foley is only making audio; this just avoids visual shimmer if you re-encode frames.

What I Ran

ocean
quiet beach shot. hint “ocean sound”. both fps at 24. it breathes with the water instead of looping flat.

fire
a tree on fire. i typed “burning tree” and ran it. close crackle, a few pops, and it tracks the flame.

beach + thunder + music
person on a small island, dark sky. i asked for “distant thunder rumbles and crackles” and added “dramatic, intense background music”. you get low thunder in the back and a simple score that fits. both at once.

car on wet road
old clip. tiny hint only. when the tires hit water, the splash lands on that frame. i didn’t line it up by hand.

windy street
strong wind, light footsteps. i barely wrote anything. it still adds wind, a soft step layer, and a little bed so it feels finished.

New Workflow for LowVram

Share This Article
Follow:
Studied Computer Science. Passionate about AI, ComfyUI workflows, and hands-on learning through trial and error. Creator of AIStudyNow — sharing tested workflows, tutorials, and real-world experiments.
8 Comments
  • I did everything as it should be, but an error pops up:
    0.0 seconds (IMPORT FAILED): E:\ComfyUI\custom_nodes\Comfyui-HunyuanFoley

  • NODE_DISPLAY_NAME_MAPPINGS File “E:\ComfyUI\custom_nodes\Comfyui-HunyuanFoley\nodes.py”, line 17 ——————————————————————————- ^ SyntaxError: invalid syntax
    And then after correction:
    Cannot import E:\ComfyUI\custom_nodes\Comfyui-HunyuanFoley module for custom nodes: No module named ‘audiotools’

  • Solved the problem: commented out line 17 in nodes.py (in ..\custom_nodes\Comfyui-HunyuanFoley). And you need to perform the operation (pip install -r requirements.txt) in the folder (ComfyUI_HunyuanFoley) with venv comfyui activated

  • Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you’re reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS=”+dynamo”

    • ComfyUI asked PyTorch to go faster (via torch.compile / “Torch Compile” nodes and via SageAttention).
      On Windows that JIT step spawns real tools:

      cl.exe & link.exe from Microsoft Visual C++ (MSVC)

      ptxas.exe from the CUDA Toolkit

      When those tools weren’t on the PATH of the process that started ComfyUI, Triton tried to run them and Windows said “I can’t find that file” → [WinError 2]

      you have to fixed this.. correctly install segattention and triton

Leave a Reply

Your email address will not be published. Required fields are marked *