WanVideo TextEmbed Bridge

WanVideoWrapper has its own text-encode node that produces its own embedding type, WANVIDEOTEXTEMBEDS. That's fine until you want to use something from ComfyUI's native conditioning world - a different text encoder, an advanced conditioning node, a combine/concat/area setup, prompt-scheduling nodes, anything that outputs standard CONDITIONING. Those two worlds don't speak the same type, so you can't just plug one into the other. This node is the adapter: it bridges native ComfyUI text embedding to WanVideoWrapper text embedding.

Plainly: it takes ordinary ComfyUI CONDITIONING and hands you WanVideoWrapper's embed type, so you can drive the wrapper's sampler with conditioning built the native way. Handy when you've got a conditioning trick you like from the core ecosystem and don't want to give it up just because you're using kijai's wrapper.

How it works

It's a type converter with intent. It reads the positive (and optionally negative) CONDITIONING tensors - the encoded prompt(s) that came out of a CLIP/UMT5 text-encode or any node that emits CONDITIONING - and repackages them into the WANVIDEOTEXTEMBEDS structure the wrapper's sampler expects. No re-encoding, no model call; it's marshaling data from one container shape into another so the downstream Wan nodes accept it.

The inputs and output that matter

• positive (CONDITIONING, required) - your positive prompt conditioning, from a native text-encode node. This is the one that has to be connected.
• negative (CONDITIONING, optional) - your negative conditioning, if you're running real CFG (i.e. CFG above 1). On the fast, speed-LoRA path where CFG is pinned to 1, the negative does nothing anyway, so you can leave it empty.
• text_embeds (WANVIDEOTEXTEMBEDS) out - wire it into the WanVideoWrapper sampler's text-embeds input in place of the wrapper's own encode node.

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. No model of its own - but whatever native text encoder you feed it does need its weights (the UMT5 text encoder for Wan) present.

Common issues & troubleshooting

Your negative prompt isn't doing anything. If you're on a speed LoRA (LightX2V / CausVid), CFG is forced to 1.0 and there's no classifier-free guidance for the negative to act through - that's a Wan-wide behavior, not this node. Feeding a negative here won't change it. If you need negative control at CFG 1, that's what the NAG-style nodes are for; the bridge just passes conditioning through.

Mismatched encoder. The CONDITIONING you feed in should come from a text encoder appropriate for Wan (UMT5). Piping in conditioning encoded for a different model family is asking for garbage or an outright error, even though the types nominally connect.

Why bother at all? If you're happy with the wrapper's built-in WanVideoTextEncode, you don't need this node - it exists specifically to let native conditioning nodes into the wrapper's pipeline. No native-side trick you care about? Skip it.