🔶 Text Prompt Clip Encode

Stock ComfyUI wants two separate CLIP Text Encode nodes for a normal generation - one for your positive prompt, one for your negative - both fed from the same CLIP loader, both wired into the two conditioning slots on KSampler. Chaosaiart's version folds that pair into a single node: one CLIP input, two text boxes, two conditioning outputs. It's the same instinct behind packs like Efficiency Nodes, which fold common multi-node sequences into fewer, denser ones - less clutter, fewer wires to trace when you're staring at a screenshot of someone's workflow.

How it works

Nothing about the actual encoding changes. Your prompt text gets tokenized and run through the CLIP text encoder exactly like the vanilla node does, producing the conditioning tensor that steers the sampler. This node runs that process twice under the hood - once per text field - it just does it inside one box instead of two. Pure graph ergonomics, not a different technique.

Inputs and outputs

• clip (required, CLIP) - wire this from your checkpoint loader's CLIP output, same as always.
• positiv_txt / negativ_txt (required, STRING, multiline) - your two prompts, side by side in one node.

Outputs: POSITIV and NEGATIV, both CONDITIONING - drop them straight into KSampler's positive and negative slots.

Installing it

ComfyUI Manager: Install via Git URL, https://github.com/chaosaiart/Chaosaiart-Nodes. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/chaosaiart/Chaosaiart-Nodes, restart. Linux users also need pip install opencv-python and pip install tqdm per the README (these serve other nodes in the pack, not this one, but they're a one-time step for the whole install). No model downloads required - this node just needs a CLIP you've already loaded.

Where people get burned

• No CLIP skip control built in. If you're running an Illustrious/Pony/NoobAI-style anime checkpoint, the community-standard advice is to set CLIP skip to 2 for correct results - in ComfyUI that means a separate CLIPSetLastLayer node (value -2) placed before this one. This node doesn't give you that knob, so you still add it upstream.
• Compression saves clicks, not compute. Bundling both prompts into one node doesn't make the encode any cheaper - it still runs twice - so don't expect a speed win, just a tidier graph.
• It's a pack-specific type, not a drop-in for shared workflows. If you're following someone else's tutorial or importing a workflow JSON built around vanilla CLIPTextEncode nodes, this won't slot into it automatically - fine for your own graphs, a minor friction if you're trying to match someone else's exactly.