VRGDG_PromptTemplateBuilder

VRGDG_PromptTemplateBuilder is where the pack turns your scattered creative notes into one structured, labeled prompt document. It gives you five sections, each with a type dropdown and a text box, and it assembles them into a single formatted prompt that downstream nodes - the LLM prompt creator, the splitters - can consume as a coherent brief.

It's the writing-desk node. Everything before this is raw material (your ideas, the lyrics, a reference image); everything after it assumes the material has been organized. Getting this node right is the cheapest way to raise the quality of the whole music video, because it's the last place you shape the creative direction before the pipeline takes over.

How it works

Five sections, and each has two widgets:

• section_N_type - a dropdown with nine options: Theme / Style, Instructions, Image to Video Prompt, Text to Video Prompt, Text to Image Prompt, Story, Lyric Segment, Ideas, and Other Notes. The type label gets written into the output so the downstream reader knows what each block is for.
• section_N_text - the actual content for that section.

Output: one formatted_prompt STRING, with all five sections assembled in order. You wire that into the LLM node or a text encoder, and the structure carries through.

How to use the nine types

Don't overthink the taxonomy, but use it deliberately:

• Theme / Style - the visual identity: "synthwave neon, late-80s VHS look." This is the high-leverage block; it colors every scene.
• Instructions - rules for the generator and LLM: "keep the character facing camera," "no text overlays."
• Story - the narrative arc, especially when you want scene-to-scene continuity (the pack's PromptSplitterJson has a summary concept for exactly this).
• Lyric Segment - the actual words being sung during this part. The workflow syncing lyrics to scenes depends on this.
• Image to Video / Text to Video / Text to Image - model-specific prompt blocks. Only fill the ones your pipeline stage needs.
• Ideas / Other Notes - freeform; the LLM reads them as context even though they're loose.

Gotchas

Empty sections aren't a problem - the builder just includes whatever you filled in. The bigger trap is stuffing an instruction into the Theme box; the downstream LLM treats the type as the semantic bucket, so "Theme / Style: no close-ups" confuses it. Keep content aligned with the label.

Installing it

Part of the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/vrgamegirl19/comfyui-vrgamedevgirl

restart, or ComfyUI Manager → search vrgamedev. No models, no dependencies beyond the pack's.

The honest take

This is one of the few nodes in the pack where your effort directly maps to output quality - it's a prompt-engineering aid, and prompt engineering is where music-video workflows live or die. Give it a tight Theme/Style block and a real Story, and the automation downstream stops sounding like a random image generator.