VRGDG Legacy Storyboard Prompt Queue

VRGDG_StoryBoardCreator is the dispatcher at the center of the pack's music-video pipeline. You hand it a list of scene prompts, and it feeds them to the rest of your workflow one at a time, keeping track of which scene you're on, and re-queuing until the whole storyboard is done. If the LLM node is the director and the group-state nodes are the stagehands, this is the stage manager calling the cues.

How it works

The prompt_list input (a multiline string, default {}) is the full set of scene prompts for your storyboard. When the node runs, it outputs the current scene's prompt, plus index, index_str, and total_prompts so the rest of the graph knows where it is in the sequence. Then auto_queue (default true) takes over: the node re-queues the workflow, and the next pass picks up the next index. The loop keeps spinning until index reaches total_prompts.

The other outputs matter for the downstream pipeline: output_folder_name and save_subpath tell the saving stages where this scene's outputs should land, so scene 3's frames don't overwrite scene 2's. That's the contract everything else in the pack depends on - a per-scene folder structure that keeps a long batch organized.

For fixing mistakes without re-rendering everything:

• use_remake_folder - re-render into a separate "remake" folder instead of clobbering the original.
• redo_mode with redo_indexes - a comma-separated list of scene indexes to redo (e.g. 2,5,7), so a bad scene 5 doesn't force scenes 1–4 to re-run.
• redo_prompt_overrides - replacement prompts for those redone scenes, in JSON-ish text form.

And trigger is the integer "advance now" pedal: bump it to force the node to move to the next index even if nothing upstream changed.

The mental model

This is where the pack's pieces connect. The LLM node writes the scene prompts → this node feeds them one at a time → VRGDG_SetGroupStateMulti swaps which group is active per scene → VRGDG_RunStateLogger_SRT logs progress → and the whole thing self-advances over hours without a human clicking. The reason music-video generation in this community is described as "leave it running overnight" is precisely this pattern.

Install and the realistic advice

Pack install: git clone https://github.com/vrgamegirl19/comfyui-vrgamedevgirl into custom_nodes, or ComfyUI Manager → "vrgamedev", then restart. Requirements kornia, librosa, imageio come with it.

If you wire this up by hand, expect prompt_list's exact format to be picky - it's a string that holds what's effectively structured data, and the pack's workflows are the best reference for what it expects. Also, auto_queue is a loop; make sure total_prompts is right before you kick it off, because "I forgot to cap the loop" on a multi-hour render is a lesson you only learn once. Test with auto_queue off and two prompts before you trust it with the whole album.