Nodes/ComfyUI-AceStep_SFT/AceStep 1.5 SFT Get Music Infos
ComfyUI Node

AceStep 1.5 SFT Get Music Infos

Analyzes audio with the native ACE-Step Transcriber to extract lyric, vocal and song-structure tags, plus BPM and key/scale via librosa. Wire outputs to Generate node or to text display nodes to inspect results.

By jeankassio·Created 5 months ago·Updated 3 months ago· 53
AceStep 1.5 SFT Get Music Infos
  • audio
  • tags
  • bpm
  • keyscale
  • music_infos
get_tagstrue
get_bpmtrue
get_keyscaletrue
max_new_tokens256
audio_duration60
unload_modeltrue
use_flash_attnfalse
temperature0.00
top_p1.00
top_k0
repetition_penalty1.10
seed0
Categoryaudio/AceStep SFT

Inputs (13)

NameTypeDefaultDescription
audioAUDIOAudio to analyze for style, BPM and key/scale.
get_tagsBOOLEANtrueExtract descriptive tags from the audio using the native ACE-Step transcriber.
get_bpmBOOLEANtrueDetect BPM from audio using librosa.
get_keyscaleBOOLEANtrueDetect key and scale from audio using librosa.
max_new_tokensoptINT25664–2000Maximum tokens for the native ACE-Step transcription output. Higher values preserve more lyric/structure detail.
audio_durationoptINT6010–300Max seconds of audio to analyze (center crop). ACE-Step Transcriber benefits from more context for sections and lyrics.
unload_modeloptBOOLEANtrueUnload the ACE-Step transcriber after use to free VRAM for generation.
use_flash_attnoptBOOLEANfalseUse FlashAttention-2 for the ACE-Step transcriber. Requires flash-attn package installed. Faster and uses less VRAM.
temperatureoptFLOAT0.000–2Sampling temperature for transcription generation. 0 = deterministic and recommended for stable structure extraction.
top_poptFLOAT1.000–1Nucleus sampling for transcription generation. Keep at 1.0 for native deterministic behavior unless you are experimenting.
top_koptINT00–200Top-K sampling for transcription generation. 0 preserves the native default behavior.
repetition_penaltyoptFLOAT1.101–3Penalty against repeated tokens in the transcription output. 1.1 is a mild safeguard against loops.
seedoptINT00–18446744073709550000Random seed for reproducible transcription generation when sampling is enabled.

Outputs (4)

NameTypeDescription
tagsSTRING
bpmINT
keyscaleSTRING
music_infosSTRING