AC_Super_CLIPEN

AC_Super_CLIPEN is the pack's answer to the two CLIP Text Encode nodes sitting at the start of every workflow. It takes a CLIP plus two plain text boxes - Positive and Negative - and hands back two ready-to-use CONDITIONING outputs. Two nodes' worth of work, one node on the canvas. Nothing more, nothing less.

How it works

Read the source and you'll find the stock encoder internals: clip.tokenize(text) then clip.encode_from_tokens(..., return_pooled=True) for each of the two strings, with the pooled output stashed in the conditioning dict the same way the core node does it. That matters for SDXL users, because pooled output is how models like SDXL get their aesthetic vector - and this node includes it, so you're not losing anything versus the stock chain. It's a faithful two-in-one, not a stripped-down version.

The inputs and outputs

• clip - a CLIP from any checkpoint or LoRA-loaded chain.
• Positive and Negative - multiline strings, with the author's default anime-style prompts prefilled. Replace them with yours.
• Outputs Positive and Negative - CONDITIONING, wired into a KSampler, a ControlNet apply node, or this pack's AC_Super_UpKSampler second pass.

The string inputs are also what make it different from AC_Super_KSampler, which encodes prompts and samples in one go. CLIPEN gives you the conditioning on its own, so you can branch it, feed it into ControlNet, or reuse it across two samplers without re-encoding.

Installing it

Part of the ComfyUI_AC_FUNV8Beta1 pack by Cc啊程 (Bilibili ComfyUI teacher; README is Chinese). ComfyUI Manager, search "AC_FUN", or:

cd ComfyUI/custom_nodes
git clone https://github.com/A719689614/ComfyUI_AC_FUNV8Beta1

Restart ComfyUI. No downloads or model files.

Where people get burned

Don't expect prompt-engineering features you won't find - there's no weighting UI, no wildcard expansion, no LoRA-stacked encoders. It's the stock encoder, just doubled up. And the defaults are a specific vibe (an anime character prompt), so your first run will look like someone else's example until you overwrite the text. One more: if you're using a model family with a separate text encoder or an exotic CLIP setup, this expects a standard CLIP object - feed it the output of your normal checkpoint loader and you're fine.