ComfyUI Node

VRGDG_RunStateLogger_SRT

A breadcrumb trail for multi-scene video renders

By vrgamegirl19·Created about a year ago·Updated about 7 hours ago· 718
VRGDG_RunStateLogger_SRT
  • trigger
  • trigger
index0
total_sets0
output_folder
note

When you're rendering a music video as a batch of scenes, each pass takes minutes on an LTX 2.3 graph, and the whole thing can run for hours. The worst failure mode isn't a crash - it's a crash three hours in, and not knowing which of the twelve scenes actually made it out. VRGDG_RunStateLogger_SRT exists to make that a non-problem: it writes a little SRT-format log of where the run is, so a resumed run doesn't re-render scenes you already have.

How it works

The node sits inline in your graph and just passes data through. It takes a trigger of type VHS_FILENAMES - that's a data type from VideoHelperSuite (the VHS pack), the sort of value its filename-returning nodes emit. This is the pack's way of saying "only fire when a video save has actually happened." You'll want VideoHelperSuite installed or this input has nothing to plug in, which is easy to miss since this node lives in a pack with an unrelated name.

Feed it the current scene's index, the total_sets of scenes in the batch, and an output_folder where the log should land. There's an optional multiline note you can drop a label or prompt fragment into - useful when you want the log to say which scene, not just that it finished. Each pass appends a line like a timestamped subtitle entry, and the trigger flows out the trigger output unchanged, so you can chain it anywhere without breaking the wire.

What you get is a checkpoint you can read with your eyes. Open the SRT, see it stopped at scene 7 of 12, and restart the pipeline with the storyboard node pointed past that index instead of hoping you guessed right.

Why this is worth it

Long AI video renders are exactly the kind of thing you walk away from and come back to. The community reports behind LTX music-video work make the pattern obvious - people describe 17-hour runs on a 12GB card, first pass at 16 steps and a second at 4. At that scale, "which scene finished" is the difference between a 20-minute resume and a 3-hour re-render. This node is a cheap insurance policy for that workflow, and honestly the kind of utility that this pack buries without fanfare.

Install and gotchas

Same as the rest of the pack:

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

or install "vrgamedev" via ComfyUI Manager, then restart. Requirements are kornia, librosa, and imageio - on Windows portable, run the requirements file from python_embeded.

Two things to actually watch: if output_folder is empty, nothing useful gets written, so set it. And the VHS_FILENAMES dependency is real - if you load a workflow using this node and Manager can't find it, the missing piece is probably VideoHelperSuite, not this pack. If you're not batch-rendering scenes, you don't need this node at all - it's pure pipeline bookkeeping, and it knows it.

CategoryVRGDG

Inputs (5)

NameTypeDefaultDescription
triggerVHS_FILENAMES
indexINT0
total_setsINT0
output_folderSTRING
noteoptSTRING

Outputs (1)

NameTypeDescription
triggerVHS_FILENAMES