ComfyUI Node
AceStep 1.5 SFT Cover / Remix
Creates the official full-song ACE-Step 1.5 Cover/Remix task. Voice and instrumental are regenerated together; singer identity is guided by Reference Audio. Branch task to TextEncode and Generate. This remains separate from the legacy latent_or_audio img2img path.
AceStep 1.5 SFT Cover / Remix
- source
- reference_audio
- task
◄source_conditioningsemantic_fsq►
◄remix_strength0.40►
◄cover_noise_strength0.20►
◄preserve_singer_identitytrue►
Categoryaudio/AceStep SFT/editing
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| source | AUDIO,LATENT | Song whose melody, rhythm, and structure will guide the cover/remix. | |
| source_conditioning | COMBO | semantic_fsq | semantic_fsq matches the normal Gradio Remix path. raw_no_fsq is the optional no_fsq path and retains more raw source acoustics. |
| remix_strength | FLOAT | 0.400–1 | Official audio_cover_strength: fraction of denoising steps that uses the source Cover context. The official musician guide recommends 0.3-0.5 for a dramatic genre change; 0.4 is the default here. |
| cover_noise_strength | FLOAT | 0.200–1 | Official Gradio Remix defaults to 0.2. It re-noises the complete source and truncates the schedule; higher values stay closer to the original recording. |
| preserve_singer_identity | BOOLEAN | true | Reuse the complete source as the official reference_audio condition so the regenerated singer keeps a recognizable timbre. The vocal performance is still regenerated in the target style. A connected reference_audio overrides this. |
| reference_audioopt | AUDIO,LATENT | Optional singer/timbre reference, identical to Reference Audio in the official Gradio UI. A clear recording of the same singer is ideal. When disconnected, preserve_singer_identity reuses the source song. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| task | ACESTEP_SFT_TASK | — |