Text Prompt

If you already know CLIP Text Encode, this node will feel almost too simple - and that's kind of the point. Text Prompt (its class name is PureText, which is a more honest description) takes a multiline string and turns it into a CONDITIONING output. No CLIP model plugged in, no tokenizer settings, nothing else to configure. It exists purely as the prompt box for ComfyUI-Attention-Distillation, the style-transfer/style-specific-generation pack built around the CVPR 2025 Attention Distillation paper.

Why it's this bare

Everywhere else in ComfyUI, encoding text into conditioning requires a separate CLIP loader wired in, because the CLIP model is a distinct component you might swap. This pack doesn't work that way - the whole pipeline (UNet, text encoder, scheduler, the works) is bundled inside the DISTILLER object that comes out of Load Distiller. PureText doesn't need its own CLIP input because it isn't doing the encoding itself; it's just handing your raw prompt text downstream to the sampler node that actually has the model loaded. That's why this node is this thin - it's a courier, not an encoder.

Practically: you'll place two of these in a style-specific text-to-image workflow, one for your positive prompt and one for your negative, both feeding ADSampler. If you're doing pure image-to-image style transfer with ADOptimizer, you won't need this node at all - that node has no text input.

The one input, and the output

• text - multiline string. The author's own tooltip is refreshingly literal: "The text to be encoded." Write a normal prompt the way you would for any SD1.5/SDXL/Flux checkpoint - this pack doesn't add special syntax on top.

Output is CONDITIONING, matching what ADSampler's positive and negative inputs expect.

Installing it

It ships as part of the whole pack - there's no way to install just this node.

ComfyUI Manager - search "ComfyUI-Attention-Distillation" and install.

Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/zichongc/ComfyUI-Attention-Distillation
cd ComfyUI-Attention-Distillation
pip install -r requirements.txt

Restart ComfyUI. You'll also need a base model loaded through Load Distiller - that's the node that actually holds the encoder this node's text eventually gets run through.

Common issues

Nothing happens if you connect it without a Load Distiller / ADSampler chain. This node on its own doesn't validate or preview anything - it just passes a string forward as conditioning. If your generation ignores the prompt entirely, check that the CONDITIONING output is actually wired into ADSampler's positive (and a separate PureText into negative), not left dangling or accidentally swapped.

Don't expect this to behave like CLIP Text Encode with weighting syntax, embeddings, or LoRA trigger-word quirks. There's no CLIP tokenizer config exposed here at all - it's a plain string being handed to whatever encoder lives inside the distiller you loaded. If you're used to (word:1.3)-style emphasis tricks from A1111/ComfyUI prompting habits, there's no guarantee this pipeline's encoder honors them the same way; treat it as a fresh start rather than assuming your usual prompting muscle memory carries over one-to-one.

Requirements install issues. This node's problems are really the pack's problems - pip install -r requirements.txt pulls in diffusers, and ComfyUI's shared-Python-environment setup means a version conflict with another installed pack is a real possibility. If the node fails to even appear after install, that's almost always where to look first, not this file.