VRGDG_MusicVideoPromptCreator

VRGDG_MusicVideoPromptCreator is the node the README tells you to customize: "Customize the Prompt Creator to define style, mood, and scenes." It takes a pile of text fields - character, lyrics, environment, lighting, camera work - and assembles them into one long, coherent scene prompt. No LLM involved, no randomness: it's a carefully built string templater, and for the LTX-2.3 music-video workflow it's the place where your creative choices actually live.

It exists because LTX-class video models reward long, detailed prompts, and because the workflow needs every scene to stay consistent. The defaults are a dead giveaway - character_description is literally "The Women." and outfit_rules is "a white dress." Change those two and you've changed the protagonist of the entire video. That's the design: set your character, style, environment, lighting, camera motion, expressions, and shot list once, feed the scene's pipe_separated_lyrics (lines separated by |), and get back a full paragraph ready for the model.

The inputs that matter

You can't realistically tune all 13 fields on every scene, so here's the short list:

• pipe_separated_lyrics - the actual lines for this scene. This is the one input that changes per scene; the rest are your video's identity.
• character_description and outfit_rules - who's on screen and what they wear. Keep these identical across every scene or your character will drift.
• word_count_min / word_count_max - target length of the assembled prompt, 10–200. Start at the 30/50 defaults; LTX likes detail but a 200-word prompt is a different beast.
• environment, lighting, camera_motion, physical_interaction, facial_expression, shots, character_visibility - the cinematic vocabulary. More specific here means fewer "the model ignored my prompt" reruns.
• signal - a wildcard trigger so the rest of the graph controls when this fires.

The single output is concatenated_string, which you wire into your prompt encoder. That's the whole node's job, and the output is deterministic - same inputs, same paragraph, every time.

How it fits the workflow

In the full pipeline, this node replaces (or complements) the LLM prompt writer: the LLM_PromptBatcher handles the story-level scene prompts, while this one is the hand-tuned, fully-controlled option for a specific scene. If you want one scene to be perfect - the chorus, the money shot - you write its lyrics in here and keep the style fields locked. It's also great standalone for hand-building LTX prompts without the whole workflow.

Setup

Same shared pack:

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

or Manager → search "vrgamedev", restart.

The one thing that bites people: the defaults are the author's demo content ("The Women.", "a white dress"), so if you run it as-is you'll get prompts about a woman in a white dress on a rain-slicked street. That's not a bug - it's a starting point you're supposed to overwrite. Miss that and every scene you generate will be inexplicably about that one outfit. Change the character fields first, then the world-building, and the node quietly becomes the best thing in the pack.