Harronode - Prompt Builder

If you've ever wrestled a text-logo prompt together by hand, you know the pain: type the word, remember the LoRA's trigger vocabulary, balance a color, a style, an accent, hope the CLIP encoder weighs the quoted text the way you wanted. Harronode (full name "Harronode - Prompt Builder") exists to kill that whole chore. It's a drop-in prompt assembler for the Harrlogos XL SDXL LoRA - the Civitai model that did "true text generation in SDXL with just a LoRA" back in late 2023 - and it puts every activation term the LoRA understands into a set of dropdowns so you never have to look them up again.

One thing up front, because it surprises people: the name is a bit of a lie. This node doesn't build prompts server-side. There's no API call, no key, no weights, no model files. The entire thing runs in your browser, and the Python class behind it is little more than a passthrough that returns whatever string sits in the prompt widget.

How it works

The pack ships a words.json with four vocabulary lists baked in: 14 colors, 18 styles, 22 accents, and 25 content items ("spikey", "dripping", "skulls", "cthulu" - you can feel the death-metal demo video). A JavaScript extension watches your widget changes and reassembles the prompt string live:

("BANDNAME":1.00) text logo, gold, 3D, fire, skulls,

Type your text in desired_image_text, set a weight, pick modifiers, and the built prompt appears in the editable prompt widget in real time. Adjust a _count widget and the node hides or reveals the corresponding color_1/2/3, style_1/2, accent_1/2/3, and content_1/2/3 dropdowns so the UI only shows what you're using. It's the rare custom node with zero Python dependencies and no requirements.txt to trip over - refreshing after a clone is genuinely all it takes.

The inputs that matter

• desired_image_text - the word or phrase you want rendered as the logo. Keep it short; this LoRA is picky about long strings.
• text_weight - 0.5 to 1.5, default 1.0. This becomes the ("TEXT":1.0) weight on the quoted text. Nudge it up if the letters wash out.
• The four *_count widgets - how many colors, styles, accents, and content terms to include. Set a count to zero and the category disappears entirely.
• The color_* / style_* / accent_* / content_* dropdowns - the actual activation terms, straight out of words.json.
• prompt - the assembled, editable result. Be aware: it's a live view. Change any parameter and it rebuilds from scratch, overwriting whatever you hand-edited. The widget's own placeholder warns exactly this.

Output is a single STRING, which you wire straight into a CLIP Text Encode node as your positive prompt.

Installing it

The README's install is refreshingly short because there's nothing to build:

cd ComfyUI/custom_nodes
git clone https://github.com/NotHarroweD/Harronode.git

then restart ComfyUI. Or skip the terminal entirely: it's in the ComfyUI Manager registry, so search "Harrlogos Prompt Builder Node" and hit install. Either way, remember the node alone does nothing - you still need the Harrlogos XL LoRA itself downloaded from Civitai and loaded with a standard LoRA loader, or you're just generating plain "text logo" prompts that no SDXL checkpoint understands.

Gotchas worth knowing

Where people get burned: this is an SDXL-era tool, in both senses. First, the ("text":1.2) weight syntax it emits is meaningful on SDXL's CLIP encoder - on the Qwen-class LLM text encoders powering newer models, those parentheses pass through as literal punctuation and do nothing. Keep this node in SDXL workflows. Second, the community buzz around Harrlogos peaked around December 2023 and went quiet by mid-2024; if your goal is just "text in the image," newer options like Flux Kontext have moved past it, but if you're already invested in the SDXL LoRA, this is comfortably the least fiddly way to drive it.

One more source-level note: the frontend writes its output by hardcoding node.widgets[17], and prompt_builder.py / harroserver.py are stubs. So if another extension injects widgets onto the same node and shifts that index, expect weirdness. None of that stops it being the right tool for the niche it fills - a genuinely zero-dependency, no-memorization prompt builder for a fun, specific LoRA.