VRGDG_MusicVideoPromptCreatorV2

Here's the real bottleneck of any AI music video: coming up with a good, distinct prompt for every scene of a three-minute song. VRGDG_MusicVideoPromptCreatorV2 is the pack's answer - a prompt-building node where you describe your character, the song's mood, a set of environments, lighting, camera moves, interactions, expressions, and shots, and it assembles all of that plus your lyrics into a single structured instruction block for every scene. The idea is that the details don't all have to be written fresh each time; you curate the palette and it composes the prompts.

It's not an LLM itself - it's a prompt assembly node (the LLM that turns these instructions into prompts sits elsewhere in the workflow). What it does is take all your comma-separated option lists and generate per-scene instructions that mix them, with a list_handling_mode controlling how the options are used.

The inputs that matter

• pipe_separated_lyrics - your lyrics as line1 | line2 | line3 (pipe-separated). This is what gets paired with the visual instructions scene by scene.
• character_description - who's in the video ("The Women." is the default, which should tell you a lot about the pack's typical output).
• song_theme_style - the overall look and mood.
• list_handling_mode - the interesting knob:
  • Strict Cycle - use each list item once, then repeat in order.
  • Reference Guide (default) - the LLM creates variations inspired by the list.
  • Random Selection - pick randomly from the list.
  • Free Interpretation - the LLM may ignore or combine items.
• environment, lighting, camera_motion, physical_interaction, facial_expression, shots, outfit_rules, character_visibility - your option lists. Defaults give you a solid cinematic starter set: open fields, neon reflections, zoom/pan/track, hair in wind, etc.
• word_count_min / word_count_max - how long each generated prompt should be (30–50 default).
• signal - an any-type trigger so the node only rebuilds when the workflow reaches it.

Output: concatenated_string - the full instruction block, ready to feed the LLM/prompt stage that turns it into per-scene prompts.

Install

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

Or search vrgamedev in ComfyUI Manager. No special dependencies - it's pure string assembly.

Common issues

• Every scene looks the same: your lists are too short or your mode is Strict Cycle with few items - the LLM keeps cycling the same options. Add more variety to the lists or switch to Reference Guide.
• Prompts too long/short: adjust word_count_min/word_count_max.
• Lyrics and visuals don't match: the node pairs lyrics with instructions, but it can't understand the lyrics. If a scene's lyric is sad and the visual list is all fireworks, that's on your curation - pick options that fit the song's arc.
• Nothing outputs: check the signal connection; the node is gated on it.

Honest take: this is a "save the boring hour" node. Writing 40 unique prompts by hand is the worst part of the music-video workflow, and this shrinks it to "curate a good palette once." It won't write prompts you'd publish for a festival - but it'll keep a long pipeline running with decent, varied scene directions, which is exactly what the assembly line needs.