Nodes/TKNodes/Smart Video Chunker
ComfyUI Node

Smart Video Chunker

Silence-based video/audio chunking for LTX 2.3 / Wan 2.2. Looks for silence in audio to create chunk breaks. This is helpful when speakers take a breath and we don't cut off speaker while talking. Chunking is required to get around VRAM issues. For Low VRAM set chunk size lower

By trashkollector·Created about a year ago·Updated 12 days ago· 1
Smart Video Chunker
  • video
  • audio
  • num_chunks
  • chunkOImages
  • chunkOfAudio
  • numberFrames
  • actual_end_time
  • actual_start_time
  • chunk_duration
  • numGenerationFrames
index0
chunk_secs10
variation2
source_fps30.00
target_fps25.00
start_time_override-1.000
model_typeLTX
CategoryTKNodes

Inputs (9)

NameTypeDefaultDescription
videoIMAGESource video
audioAUDIOSource audio
indexINT00–9999Index from Loop - zero based
chunk_secsINT10Size of each video segment in seconds
variationINT2Num seconds variation. chunks_secs +/- variation adds flexiblity to find silence
source_fpsFLOAT30.001–240TRUE fps of incoming video tensor
target_fpsFLOAT25.001–240fps required by target model - 25 for LTX, 16 for WAN (typical)
start_time_overrideoptFLOAT-1.000-1–999999Use -1 for index 0, used to maintain exact timing of chunks.
model_typeoptCOMBOLTXTarget model frame-count boundary: LTX=8n+1, WAN=4n+1

Outputs (8)

NameTypeDescription
num_chunksINTNumber Chunks Calculated for the Video.
chunkOImagesIMAGEVideo
chunkOfAudioAUDIOAudio
numberFramesINT# Frames after Snapping
actual_end_timeFLOATend time in video of chunk
actual_start_timeFLOATstart time of chunk in video
chunk_durationFLOATlength of chunk
numGenerationFramesINTnum Frames requested by user w/o snapping