Nodes/ComfyUI-JK-HeartMuLa/HeartMuLa Style Embed
ComfyUI Node

HeartMuLa Style Embed

A ComfyUI node in JK-HeartMuLa with 7 inputs and 1 output.

By Crono141·Created 3 months ago·Updated 3 months ago· 0
HeartMuLa Style Embed
  • muq_model
  • audio_input
  • cmuq
enabletrue
free_vram_aftertrue
normalizetrue
audio
style_strength1.00
CategoryJK-HeartMuLa

Inputs (7)

NameTypeDefaultDescription
muq_modelJKHEARTMULA_MUQ
enableBOOLEANtrueWhen off, output a zero embedding (no style transfer) -- identical to style_strength = 0. Lets a single switch toggle style transfer on/off in a shared workflow without rewiring.
free_vram_afterBOOLEANtrueAfter embedding, offload MuQ-MuLan back to CPU and free its VRAM so the HeartMuLa models have room to generate. No effect when MuQ runs on CPU. It stays cached in RAM (no reload), and is moved back to the chosen device automatically on the next embed.
normalizeBOOLEANtrueRe-normalize the reference embedding to unit length before applying style_strength, so the strength is consistent regardless of reference length. (MuQ averages 10s clips, so a longer reference otherwise yields a weaker vector.) Recommended on.
audioCOMBOReference audio (drag-drop or upload). Ignored if the 'audio_input' socket is connected.
style_strengthFLOAT1.000–2Style influence. 0 = off (no style), 1.0 = natural/calibrated, >1 over-drives (can destabilize -- short/subdued output). Drag or type a value.
audio_inputoptAUDIO

Outputs (1)

NameTypeDescription
cmuqJKHEARTMULA_CMUQ