WanVideo Combine Embeds

Small node, single job: take two sets of Wan image embeds and combine them into one. If you've built up conditioning from two different sources - two context windows, two init segments, two chunks you want the sampler to treat as one continuous sequence - this welds them together so the rest of the graph sees a single WANVIDIMAGE_EMBEDS.

It's connective tissue for the more advanced Wan workflows, especially anything doing long-form generation by stitching windows. On a plain single-image I2V or text-to-video graph you'll never need it. It earns its place the moment you're juggling more than one embed stream - two separate context windows you want sampled as one clip, an intro segment prepended to a main segment, or embeds built from different init sources that need to become a single sequence before they hit the sampler.

How it works

WANVIDIMAGE_EMBEDS is the wrapper's container for latent image conditioning. This node concatenates two of those containers into one - embeds_1 followed by embeds_2 - so their frames sit end-to-end in a single conditioning object. The sampler then generates against the combined sequence as if it were built in one pass. There's no blending or averaging here; it's a join, not a mix.

The inputs and output that matter

Both inputs are required and there are only two:

• embeds_1 (WANVIDIMAGE_EMBEDS) - the first set.
• embeds_2 (WANVIDIMAGE_EMBEDS) - the second set, appended after the first.

Order matters - embeds_1 comes first in the resulting sequence. The output is image_embeds (WANVIDIMAGE_EMBEDS), which you hand to the sampler or the next embed-modifying node.

Want to combine more than two? Chain the nodes: feed the output of one WanVideoCombineEmbeds into embeds_1 of the next, and so on.

How to install it

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 extra to download.

Common issues & troubleshooting

Mismatched embeds. Both inputs should be the same kind of Wan image embeds - same resolution, same construction. Combining embeds built for different resolutions or from incompatible sources is asking for a shape error or, worse, a subtly wrong generation. Keep the two branches consistent.

Order confusion. If your combined clip has the two halves in the wrong sequence, swap which set goes into embeds_1 vs embeds_2. There's no reverse toggle - the join is simply first-then-second.

It's not a long-video solution by itself. Combining embeds is one piece of window-stitching, not the whole thing. The 81-frame native context and identity drift across chunk boundaries are unchanged - the community's single-node chunkers (SCAIL-2 Infinity, scail-auto-extend, Bernini Infinity) exist precisely because doing this by hand is fiddly. If you're building long-form from scratch with combine nodes, expect to also manage overlap and re-anchoring yourself.