CogVideo TextEncode Combine

Sometimes one prompt isn't enough and you want to mix two. This node takes two already-encoded conditionings and merges them into a single one the sampler can use. Think prompt blending - half "a serene mountain lake" and half "golden hour lighting," fused into one conditioning rather than crammed into one text box. It's a small, focused utility that sits downstream of two CogVideo TextEncode nodes.

Why blend at the conditioning level instead of just writing a longer prompt? Because you get control over the mix. A single prompt gives every clause whatever weight the model decides; combining two encoded prompts lets you set the ratio explicitly, or concatenate them so both are fully present. It's the difference between hoping the model balances your ideas and telling it how.

The inputs that matter

• conditioning_1 and conditioning_2 - the two CONDITIONING outputs you're merging, each from its own TextEncode node.
• combination_mode (default weighted_average) - how they blend:
  • weighted_average - mixes the two by a ratio you set. The flexible default.
  • average - a straight 50/50 blend.
  • concatenate - appends them so both prompts are fully represented rather than averaged. Use this when you don't want either idea diluted.
• weighted_average_ratio (default 0.5) - the mix balance for weighted-average mode. 0.5 is even; push toward 0 to favor the first conditioning, toward 1 to favor the second.

The single output, conditioning, goes into the sampler's positive (or negative) input, exactly where a normal single conditioning would.

How to install it

Part of the wrapper:

• ComfyUI Manager - search ComfyUI-CogVideoXWrapper, install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/kijai/ComfyUI-CogVideoXWrapper, then pip install -r ComfyUI-CogVideoXWrapper/requirements.txt, restart.

Common issues & troubleshooting

The blend leans too hard one way. Adjust weighted_average_ratio. It's easy to forget which conditioning is 1 and which is 2 - if the result favors the wrong prompt, either flip the ratio or swap the inputs.

Averaging muddies both ideas. Averaging two conditionings can wash both out into something generic. If you want both concepts present and distinct, switch combination_mode to concatenate instead - it keeps them whole rather than splitting the difference.

Do I actually need this? Often not. For most prompts, one well-written TextEncode is plenty - this model takes long descriptive prompts happily, so you can usually just write both ideas into a single prompt. Reach for Combine when you specifically want dial-able control over the balance between two separate prompts, or for prompt-transition experiments.