Manual Lyrics Extractor (SRT Segments)

VRGDG_ManualLyricsExtractor_SRT is the "you brought your own lyrics file" path into the music-video workflow. Give it an SRT subtitle file containing your lyrics with timestamps, connect the actual audio, and it converts those timestamped lines into scene-sized lyric segments the rest of the pipeline can use. The word "Manual" is the honest part - instead of transcribing your song for you, it trusts the SRT you already have.

That's the typical setup with this pack: people write their songs in Suno, then feed the lyrics as an SRT to the workflow. The subtitle file carries the timing information the video needs - you can't sync scenes to a song without knowing when each lyric lands.

The inputs that matter

• srt_path - path to your .srt file. This is the lyrics plus timestamps.
• audio - the actual audio track (an AUDIO input). The node uses it for the real duration and timing, so the segments match the song rather than a guess.
• scene_duration_seconds - how long each scene should be, default 4. This sets the granularity: shorter scenes, more of them; longer scenes, a slower-paced video.
• fps - 25 default. Used to convert SRT timecodes to frames, since downstream video nodes think in frames.
• use_ltx2 - off by default. Flip it on when you're targeting LTX-2/2.3, which have their own frame-count and timing constraints (frames divisible by 8 plus 1) - the node adjusts its segment math to play nice with those.
• language - defaults to english, with a dropdown spanning 113 options including auto. This sets the language context for how the lyrics are treated downstream.

The single output, all_lyrics_combined, is the lyrics stitched together with their scene timing - wire it into the prompt-building stages (like VRGDG_MusicVideoPromptCreator or the LLM batcher) so each scene gets written against what's actually being sung.

Getting it set up

Shared pack install:

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

or Manager → search "vrgamedev", restart. Audio handling in ComfyUI needs an audio node - VHS or a core audio loader - plus the pack's librosa dependency if you're doing any audio math.

Where people get stuck

The most common failure is pointing srt_path at a file that doesn't exist or isn't actually an SRT, and then wondering why nothing comes out - the node won't guess your file location for you. Also: if your SRT timestamps don't roughly match the audio you connected, your "synced" scenes will be off by whatever the mismatch is, because the node trusts the file. And use_ltx2 isn't cosmetic - leave it off for non-LTX targets and on for LTX-2/2.3, or you'll fight the model's frame constraints. Get those three right and this node quietly does the tedious timestamp work for you.