Dual Prompt CLIP Encode

What it is

If you've ever watched the DeepSeek Illustrious Prompt generator spit out a perfect English tag list and then had to copy it by hand into the standard CLIP Text Encode node, this is the node that ends that ritual. Dual Prompt CLIP Encode takes the generator's positive_prompt and negative_prompt strings, encodes both into CONDITIONING in a single node, and hands the result straight to the KSampler. It's the middle link in the pack's recommended chain:

DeepSeek Generator → Dual Prompt CLIP Encode → KSampler

How it works

Under the hood there is no mystery at all: it calls clip.tokenize() and clip.encode_from_tokens_scheduled() on each prompt - the exact same path a normal CLIP Text Encode node uses, just for two prompts at once and with forceInput enabled so the strings can be wired in rather than typed. That's the whole trick, and it's a good one: the node itself contains no model, no API call, and no hidden behavior, so the encoding you get is byte-for-byte what you'd get from the boring default nodes. You're paying for convenience, not for a different result.

The inputs and output

• clip - from your Load Checkpoint. Works with any SDXL-lineage CLIP, which covers Illustrious, NoobAI and the merge layer that grew on them.
• positive_prompt - wire the generator's English positive_prompt here.
• negative_prompt - wire the generator's negative_prompt.

It returns positive and negative CONDITIONING, which go into the KSampler's positive and negative slots. Two prompts in, two conditionings out, nothing else.

Gotchas worth knowing

The one mistake that bites people: feed it the English fields, not the Chinese one. positive_prompt_chinese is a translation for you to read, and feeding Chinese text into an Illustrious CLIP that was trained on Danbooru tags will give you mush. It's not an error - it'll happily encode it - but the tags it encodes are the ones the checkpoint understands.

Also remember this node is only meaningful inside this pack's flow. On its own it's just a tidier CLIP Text Encode; there's nothing here that makes your prompting better. And because the generator is a cloud call, this node can't do anything until the generator has run - encode after, not before. Install it the same way as the rest of the pack (Manager, or git clone the plugin into custom_nodes and restart); it adds no dependencies and downloads nothing.