ACE-Step Understand

A ComfyUI node in Audio/ACE-Step with 12 inputs and 6 outputs.

By kana112233·Created 6 months ago·Updated 6 months ago· 27

ACE-Step Understand

◄checkpoint_dirAce-Step1.5►

◄config_pathacestep-v15-turbo►

◄lm_model_pathacestep-5Hz-lm-1.7B►

◄target_duration30.00►

◄deviceauto►

◄languageauto►

◄temperature0.3►

◄top_k0►

◄top_p0.90►

◄repetition_penalty1.0►

◄thinkingtrue►

CategoryAudio/ACE-Step

Inputs (12)

Name	Type	Default	Description
audio	AUDIO		The audio signal to be analyzed.
checkpoint_dir	COMBO	Ace-Step1.5	Directory containing ACE-Step model weights (DiT model).
config_path	COMBO	acestep-v15-turbo	Specific model configuration to use (e.g., v1.5 turbo).
lm_model_path	COMBO	acestep-5Hz-lm-1.7B	Path to the language model used for audio analysis and understanding.
target_duration	FLOAT	30.0010–600	Target duration to reference during analysis.
device	COMBO	auto	Computing platform to run the model on.
languageopt	COMBO	auto	Hint the model about the vocal language in the audio.
temperatureopt	FLOAT	0.30–2	Sampling temperature. Lower (0.0-0.3) = more precise/faithful, Higher (0.5+) = more creative. Try 0.1 for better accuracy.
top_kopt	INT	00–100	Top-K sampling. 0 = disabled. Lower values (e.g., 20-50) can improve accuracy by limiting token choices.
top_popt	FLOAT	0.900–1	Top-P (nucleus) sampling. 1.0 = disabled. Lower values (e.g., 0.8-0.9) can improve accuracy.
repetition_penaltyopt	FLOAT	1.00.5–2	Repetition penalty. 1.0 = no penalty. Higher values (1.1-1.3) reduce repetitive lyrics.
thinkingopt	BOOLEAN	true	Whether to show the language model's Chain-of-Thought reasoning.