Nox Prompt Palette Mixer

Color is the cheapest way to set a mood and the easiest thing to get wrong. Nox Prompt Palette Mixer is the pack's color node: pick a mood_profile and an intensity, and it outputs palette override strings, custom keyword bundles, and usage notes. It's one of the lightest nodes here - five inputs total - and it's designed to feed the Builder, whose palette_overrides and custom_keywords inputs are named after exactly what this node outputs. The README's trio (Narrative Weaver → Palette Mixer → Builder) treats it as the color half of the storytelling stack.

How it works

Curated mood profiles in constants.py map a cinematic mood to palette language: the override strings describe the color treatment ("warm golden grade, honeyed skin tones, amber highlights") and the keyword bundle provides compact terms the Builder can slot into its keyword sections. intensity scales how strongly that palette reads, include_effects toggles whether effects language (bloom, glow, grain) rides along, and custom_palette is free-text if you want to hand-write your own color language and skip the vocabularies entirely.

Inputs and outputs that matter

Required: mood_profile and intensity. That's the whole decision surface - pick a mood, set how loud it should be. Outputs: palette_overrides (wire into the Builder's palette_overrides input), custom_keywords (into the Builder's custom_keywords), notes (the readable explanation for your own review), and preset_status. Because there are only two real knobs, preset_action/preset_name lets you save moods you like as one-line recalls.

Installing it

Pack install, once. ComfyUI Manager: search "Nox Prompter", install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/jamesk9526/NoxPrompter-ComfyUI-CustomNode

Restart, find the NoxPrompter category. No dependencies, no models.

Gotchas

Standard stale-README note (use the URL above, not the old ComfyNoxPrompter one). The more useful warning: palette overrides are descriptive color language, so they work best on models that respond to sentence-level color prose - on tag-based SDXL checkpoints the effect is weaker, and on LLM-encoded models you want them phrased as instructions, which the curated vocabularies mostly are. Also, if you feed the same palette into both the Builder and a separately-authored scene prompt, you'll get doubled color terms; pick one owner for color, same as the pack's centralized-lighting rule.