ACE-Step 1.5 cover workflows still miss originals

ACE-Step 1.5's cover mode is real, but it still feels more like experimental style transfer than a dependable song-reconstruction tool; community reports suggest the best results come from tiny lyric edits, careful prompt shaping, and keeping the source song close to the target arrangement; ComfyUI's implementation appears to lag the native ACE-Step 1.5 experience, with users calling out missing adherence sliders and brittle duration handling; when the model misses, it tends to lose the original melody/chord DNA and collapse into generic pop instead of a faithful cover; LoRA-tuned variants and source-specific workflows seem to matter more than swapping random public JSONs.