WanVideo Empty MMAudio Latents

This is a tiny utility node with a specific home: Ovi, the joint audio-and-video generation pipeline. Wan has no native audio - the KB is clear that adding sound means a separate stage - and Ovi is one of the answers, a model that generates video and its matching audio together. This node makes the empty audio latent that the Ovi pipeline fills in. Think of it as the "Empty Latent Image" node, but for the audio track.

You'll notice its category is WanVideoWrapper/Ovi - that's your signal it only makes sense inside an Ovi graph. Outside of one, there's nothing for the audio latent to feed.

How it works

Diffusion generates by denoising from a blank starting latent. A joint audio-video model like Ovi needs two blank canvases: one for the frames, one for the sound. This node produces the audio one - an empty MMAudio-format latent of the right size for your clip. The sampler then denoises it into an actual audio track in lockstep with the video, so the two come out synchronized rather than bolted together afterward.

The single input is length, and it needs to correspond to your video length so the audio and picture line up.

The inputs and outputs that matter

• length (default 157) - the size of the audio latent, which sets how long a track gets generated. This must match up with the frame count / duration of the video side of your Ovi job. The default corresponds to Ovi's standard clip length; if you change your video length, change this to match or the audio and video won't be the same duration.

Output: samples (LATENT) - the empty audio latent, into the Ovi sampling path.

How to install it

ComfyUI Manager → search ComfyUI-WanVideoWrapper, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/kijai/ComfyUI-WanVideoWrapper
pip install -r ComfyUI-WanVideoWrapper/requirements.txt

then restart. This node downloads nothing itself, but the Ovi workflow it belongs to needs the Ovi model weights plus the audio components on disk. Ovi generates audio and video jointly, so it's heavier than plain Wan - budget VRAM accordingly and grab an fp8/quantized build if you're not on a big card.

Common issues & troubleshooting

Audio and video are different lengths / drift apart. This is the one thing to get right. length here has to correspond to your video frame count. Leaving it at the default while rendering a longer or shorter clip is the classic desync. Set them together.

The node does nothing on its own. Correct - it's an input generator for Ovi. If you dropped it into a normal Wan graph expecting audio, that's the confusion: it only feeds the Ovi audio path. Build the full Ovi workflow around it.

Node missing / lives under a submenu. It's filed under WanVideoWrapper/Ovi in the node menu, so it won't show up where the other WanVideo nodes are - look under the Ovi submenu. If it's absent entirely, your wrapper predates Ovi support; update and restart.

Everything OOMs. Joint audio-video is demanding. Reduce resolution, use a quantized Ovi build, and keep other models offloaded - you're running two generative processes at once.