WanVideo Preview Embeds

WANVIDIMAGE_EMBEDS is the wrapper's catch-all container for image/video conditioning - VACE encode outputs one, UniLumos embeds outputs one, the ReCamMaster camera embed outputs one. It's opaque by design: a bundle of latents and masks packed together so downstream nodes don't need to know which upstream node built it. Which is convenient right up until something looks wrong in your generation and you want to know what's actually inside that bundle before you burn GPU minutes finding out the hard way. That's what this node is for.

What it does

It's a debug tap. Feed it an embeds bundle, and it unpacks the pieces you can actually look at: the image latent inside it, and the mask. Nothing is transformed or recomputed - it's reading out what's already there.

The inputs and outputs that matter

• embeds (WANVIDIMAGE_EMBEDS, required) - any embeds bundle from anywhere in the wrapper: VACE encode, image-to-video encode, UniLumos, a camera embed, whatever you're trying to inspect.
• image_embeds (LATENT) - the latent tucked inside the bundle. Wire it into a VAEDecode (with the matching Wan VAE) and a PreviewImage to actually see it as pixels.
• mask (MASK) - whatever mask, if any, was packed alongside the latent. Wire it into a mask preview node the same way.

Neither output feeds back into your generation graph - this is a side-branch you tap off your main pipeline for a sanity check, then discard.

How to install it

Ships with the pack. ComfyUI Manager: search ComfyUI-WanVideoWrapper, install, restart. Manually:

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

then restart. Nothing to download for this node specifically - it's pure plumbing.

Common issues & troubleshooting

The decoded preview looks like noise or garbage colors. Make sure you're decoding with the Wan VAE that matches whatever produced the embeds in the first place. A 2.2 5B latent decoded through the wrong VAE variant, or vice versa, will look wrong even though nothing about the embeds themselves is broken.

The mask output is blank or solid. Not every embeds bundle carries a meaningful mask. A plain image-to-video encode might pack an all-ones (or empty) mask because there's nothing to mask - that's expected, not a bug. Masks get interesting on VACE or UniLumos embeds, where masking is actually part of the conditioning.

Nothing seems to change when you tweak upstream settings. Remember this node only shows you what's already baked into the bundle at the moment you tap it. If you change a parameter upstream and don't see it reflected, check you're not caching an old embeds object somewhere in the graph - re-run the upstream node so the bundle you're previewing is current.

Where this actually earns its keep: anything with a control video, reference images, or a foreground/background split - VACE, UniLumos - where it's genuinely hard to eyeball whether your inputs got packed the way you expect. Drop this node in, look at the latent and mask before you commit to a full sampling run, and save yourself a wasted generation.