🧩LC Prompt Assembler

This is the payoff node of the pack's whole Prompt Builder stack. Feed it the outputs of the modular prompt nodes - subjects, scene, camera, lighting, style, palette - and 🧩LC Prompt Assembler returns two things: a clean, ordered prompt string for normal CLIP encoding, and a structured JSON with bounding boxes for the regional models that want placement data (Krea2, Ideogram, the Anima regional tooling). One prompt, two consumers, built from parts instead of one giant box of text.

How it works

All the content inputs are optional sockets, each with a tooltip naming its source: subjects (from 🗒️LC Subject or Subject Array), scene (Scene Builder), camera, lighting, style, palette (Color Palette's text output), plus free-text mood, background, composition, and extra fields. It assembles the human-readable prompt with labeled sections - Subject, Scene, Camera, Lighting, and so on - and, separately, builds the JSON for regional use.

The JSON part is where it gets clever. Subject nodes carry position trailers (which this pack encodes as |||LC_POS:h|v|d||| markers), and the assembler converts those into elements entries with a bbox - a frame-relative box derived from horizontal, vertical, and depth placement, inset from the edges by margin (0–0.25, default 0.05 = 5%). Scene furniture/landmark boxes are included only when include_scene_bboxes is on (off by default - the main regional use is subject boxes). If there's high-level content but no placed elements, it emits a single centered fallback box rather than nothing.

The three widgets that matter:

• include_scene_bboxes - off by default; flip it only if you want scene objects in the JSON.
• remove_empty - skip blank sections so empty sockets don't leave stray lines.
• pretty_print - indented vs compact JSON. For humans or for log files.

Outputs: prompt (STRING → CLIP text encode / conditioning) and json (STRING → regional builders like Krea2 or Ideogram).

The real workflow

Subjects + Scene + Camera + Lighting + Style + Palette → Assembler → prompt to CLIP for the full image, json to a regional builder for placement-aware generation. The pack positions the json output as only for regional builders - don't feed it to CLIP and expect magic.

Install

Part of ComfyUI LC123 Nodes (MIT, 89 nodes). ComfyUI Manager → search "ComfyUI LC123 Nodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/lonecatone23/ComfyUI_LC123_nodes

Restart. No extra dependencies. On hosted platforms like RunningHub, request the pack and they install it for you.

The honest take

The prompt side of this node is genuinely nice - block-structured prompts are exactly what modern LLM-encoded models respond to. The JSON side is only as good as your regional model's appetite for bbox captions; Krea2 and Ideogram are the intended consumers, and if you're not using those, the json output is just neat decoration. Start with the prompt, add the regional part when the model asks for it.