Prompt (full-pipe)

Prompt (full-pipe) is the pack's big prompt box: it takes a FULL_PIPE in, reads the model and CLIP from it, and writes conditioned positive/negative back into the pipe. On the surface it's just two text fields that become conditionings. Underneath it's also a LoRA loader, a quality-tag injector, a style switcher, a character preset matcher, and a wildcard roller - which is why the author calls the pack "nodes to clean up your workflow." This is the node that replaces about four others in a typical anime workflow.

You reach for it when your workflow has outgrown the raw CLIPTextEncode and you keep copy-pasting score_9, score_8_up into the positive box by hand. It was built for the SDXL/Danbooru-tag world - the docs are explicit that the preset system targets SDXL and that Anima (with its stricter prompt ordering) is better served by Simple Prompt instead.

What it does

Feed it the pipe and a positive/negative string and it:

• Strips # comment lines - handy for keeping a stash of alternate prompts in one box.
• Extracts <lora:file.safetensors:strength> syntax and applies the LoRAs to the model and CLIP in the pipe for you. A third colon gives separate model and CLIP strengths: <lora:foo:0.8:0.6>.
• Adds quality tags and embeddings from the model preset (toggle with quality_tags and embeddings), plus a style preset (anime/realistic), character presets, tag presets, and wildcard text if you've defined them in the Preset Manager.
• Splits the result into separate conditionings per source (quality, style, character, main prompt) and concatenates them, instead of jamming everything into one giant string. That's a real improvement over a plain text encoder - the model sees coherent chunks rather than a comma soup.

The inputs that matter

The two you'll touch every run are positive and negative (both multiline). The toggles that matter:

• quality_tags and embeddings - on by default. If your checkpoint isn't Pony/Illustrious, or you're getting tags you didn't write, these are the first things to turn off.
• style - dropdown of your style presets, none by default.
• mode - concatenate (default), combine (averages the conditionings), or join (encodes everything as one string). Leave it on concatenate unless you know why you want otherwise.
• trigger_words - a force-input string, so you can wire a separate node's output in as a trigger tag source.
• deduplicate_tags - on by default; removes repeated tags from the final text.

Output is the updated full_pipe, now carrying conditioned positive/negative, the LoRA-patched model/clip, and a meta dict with the prompt text - which the Save node uses to write A1111-style metadata.

Installing

Part of the mudknight utils pack - Manager, search comfyui-mudknight-utils, or:

cd ComfyUI/custom_nodes
git clone https://github.com/mudknight/comfyui-mudknight-utils

Restart ComfyUI. No extra downloads for this node, though the preset features shine if you open the Preset Manager (the button the pack adds next to the Manager button) and define models, styles, characters, and tags.

Common gotchas

• Prompt is empty because of a # - any line whose first character is # is dropped. That's intended, but it bites people who paste a hashtag-heavy prompt. Escape with \# if you really need a literal one.
• Mystery quality tags - you've got a Pony or Illustrious preset matching your checkpoint name (the model preset matches by filename, so waiIllustriousSDXL_v160.safetensors maps to its own family). Turn quality_tags off if you don't want them.
• Negative weights don't work - this node doesn't add negpip; that's the Simple Prompt node's toggle. If you want negative weights, swap.
• Anima users, use Simple Prompt - the author says the preset machinery's prompt ordering conflicts with Anima's requirements.

It's a niche pack with zero comfy.icu impressions, so you found this through a shared workflow or the docs site - but if you're building a Danbooru-tag SDXL pipeline, this node genuinely removes a lot of graph clutter.