Nodes/ComfyUI-QwenASR/Subtitle (QwenASR)
ComfyUI Node

Subtitle (QwenASR)

A ComfyUI node in 🧪AILab/🎙️QwenASR with 15 inputs and 4 outputs.

By 1038lab·Created 6 months ago·Updated 6 months ago· 65
Subtitle (QwenASR)
  • audio
  • TEXT
  • SUBTITLES
  • LANUGAGE
  • OUTPUT_PATH
â—„modelQwen/Qwen3-ASR-0.6Bâ–ş
â—„precisionbf16â–ş
â—„attentionautoâ–ş
â—„forced_alignerQwen/Qwen3-ForcedAligner-0.6Bâ–ş
â—„languageautoâ–ş
â—„hintsâ–ş
â—„output_formatnoneâ–ş
â—„output_pathâ–ş
â—„split_modesplit_by_punctuation_or_pause_or_lengthâ–ş
â—„max_gap_sec0.6â–ş
â—„max_chars40â–ş
â—„max_inference_batch_size32â–ş
â—„max_new_tokens256â–ş
â—„unload_modelstrueâ–ş
Category🧪AILab/🎙️QwenASR

Inputs (15)

NameTypeDefaultDescription
audioAUDIOAudio input to transcribe.
modeloptCOMBOQwen/Qwen3-ASR-0.6BChoose the ASR model size.
precisionoptCOMBObf16Inference precision.
attentionoptCOMBOautoAttention backend override.
forced_aligneroptCOMBOQwen/Qwen3-ForcedAligner-0.6BForced aligner for timestamped subtitles.
languageoptCOMBOautoForce language or auto-detect.
hintsoptSTRINGOptional hints/keywords (names, terms) to improve recognition.
output_formatoptCOMBOnoneFile save format only (does not change subtitle output).
output_pathoptSTRINGOptional output file path (relative goes to ComfyUI output).
split_modeoptCOMBOsplit_by_punctuation_or_pause_or_lengthSentence splitting strategy.
max_gap_secoptFLOAT0.60–8Max silence gap to keep the same sentence.
max_charsoptINT400–200Optional max characters per line (0 = no limit).
max_inference_batch_sizeoptINT321–256Batch size for inference/alignment to avoid OOM.
max_new_tokensoptINT2561–2048Max new tokens per chunk.
unload_modelsoptBOOLEANtrueUnload cached model after inference.

Outputs (4)

NameTypeDescription
TEXTSTRING—
SUBTITLESSTRING—
LANUGAGESTRING—
OUTPUT_PATHSTRING—