VRGDG_LoadAudioSplit_Wan22HumoFMML

This is the most "automated" of the pack's audio splitters, and it's built for the Wan-Humo side of the VRGameDevGirl music-video factory. VRGDG_LoadAudioSplit_Wan22HumoFMML slices your audio into scene-length chunks - but unlike the plain upload splitter, it also manages the queue: it counts how many scenes are already done from the output folder, computes the remaining sets, and can auto-queue the next batch of generations. It's less "give me chunks" and more "keep the pipeline running until the song is done."

The mechanism: you feed it audio and a trigger signal, and it splits at a fixed scene_duration_seconds (1–5 seconds, default 4). It computes frames_per_scene at 25 fps, adjusts the count for the Humo model's frame constraints, and derives index, total_sets, and groups_in_last_set - the same bookkeeping the combine stage needs. The index is derived by scanning your folder_path for already-written *-audio.mp4 files, so restarts pick up where they left off instead of redoing finished scenes.

The inputs that matter

• audio - your AUDIO tensor.
• trigger - an any-type signal that gates execution; wire whatever marks "start" in your workflow.
• scene_duration_seconds - 1 to 5, default 4. How long each scene's audio chunk is. This one's capped low because it's tuned for short Humo scenes.
• folder_path - where output videos go (default video_output); used to count already-done scenes.
• enable_auto_queue - on by default. True lets the node auto-queue the remaining sets of scene generations.

Outputs are the full automation kit: meta, total_duration, index, start_time, end_time, instructions, total_sets, groups_in_last_set, frames_per_scene, audio_meta, output_folder, audio_1..audio_16, and a signal_out to chain the next stage. Everything the combine/trim nodes need to stay in sync.

Install

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

Or search vrgamedev in ComfyUI Manager. It's part of the pack's Humo automation module (HumoAutomation.py); if the module fails to import, the usual culprit is a missing dependency from requirements.txt, so let Manager install them or pip install -r requirements.txt.

Common issues

• It starts over from scene 0 every time: the index comes from counting *-audio.mp4 files in folder_path. If your output folder doesn't have the expected naming, the count is zero and it restarts. Keep the folder layout consistent.
• Scenes cut mid-beat: scene_duration_seconds is a fixed window, not beat-synced. If you need musical cuts, this node isn't the tool - use a transcribe-then-split flow and set real durations.
• Auto-queue surprises: with enable_auto_queue on, the node queues work automatically. If you want to review before generating, turn it off - it's the one setting in this node that can make your GPU busy without you clicking anything.

Honest take: it's the most opinionated splitter in the pack - it assumes you're running the full Humo music-video assembly line and want it to keep going. If you are, it's genuinely handy (the "resume from folder count" behavior is a small thing that saves real pain on long songs). If you just want chunks, the simpler VRGDG_LoadAudioSplitUpload is less surprising. Both are valid; they're built for different people.