Created this test workflow with MMAUDIO. It can be found in Comfy Manager. Here are the models, vae, and everything else you'll need. <a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/MMAudio_safetensors/tree/main">https://huggingface.co/Kijai/MMAudio_safetensors/tree/main</a> Here's the GitHub just in case. <a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-MMAudio/tree/main/mmaudio">https://github.com/kijai/ComfyUI-MMAudio/tree/main/mmaudio</a> This is new and still experimental so expect weird things to happen. 2 VERY IMPORTANT THINGS TO DO IF YOU ARE ON ANYTHING NOT NIVIDIA. I got this workflow working by changing a few things. I run on Intel Arc B580 so no cuda for me. I was able to make some changes to two files that let me use this with cpu. In ComfyUI\custom_nodes\comfyui-mmaudio\<a target="_blank" rel="ugc" href="http://nodes.py">nodes.py</a> I changed line 77 and 78 to<pre><code>clip_frames = torch.stack([clip_transform(frame.cpu()).to("cpu") for frame in clip_frames])
sync_frames = torch.stack([sync_transform(frame.cpu()).to("cpu") for frame in sync_frames])
</code></pre>In ComfyUI\custom_nodes\comfyui-mmaudio\mmaudio\ext\autoencoder\<a target="_blank" rel="ugc" href="http://vae.py">vae.py</a>I changed the four spots where cuda was mentioned to cpu.It has worked on a test for me but I don't generate long videos so not sure how much it can do.

ComfyUI_INTEL_LOWVRAM_WAN2.2_ADD_AUDIO

MMAudio with pre generated Wan2.2 vid.

Files with SHA256: 8265bb6de9832ffae2481fbb78067180e078c8f71e9d0edd37390382f7ae8299