Improve a song you already like
First identify whether the problem is a word, a musical passage, or the mix. Preserve the source and choose a focused revision rather than assuming a new render will keep everything unchanged.
First identify whether the problem is a word, a musical passage, or the mix. Preserve the source and choose a focused revision rather than assuming a new render will keep everything unchanged.
1. Describe the audible problem
| What you hear | First check | Possible next step |
|---|---|---|
| A wrong word or crowded lyric | Compare the sung phrase with the submitted text. | Clarify the lyric or phrase; consider a section edit or new generation. |
| The melody is wrong | Check the recognized or edited reference notes. | Correct the musical input before generating again. |
| The lead is hidden behind instruments | Compare the arrangement in context. | Try a clearer production brief, or assess stems for separate mixing. |
| Noise, crackle, or a watery separated vocal | Compare the original file and each processing stage. | Fix the source problem or choose a cleaner take rather than adding more processing. |
2. Preserve the original and make a focused change
Keep the original song and its input text. Note the timestamp and write one clear goal, such as “make the last word of the chorus understandable.” If the whole arrangement needs to change, revise the relevant Style or reference direction and generate a new candidate.
Do not add several conflicting descriptions such as “very intimate” and “huge crowded choir” unless that contrast is intentional. Repeating the same request is not a guarantee of a better result, and each request uses its applicable generation cost.
3. Use a section edit when the task fits
For the documented ACE-Step route, open Create → Custom, choose ACE-Step, and add one source song through Audio or From library. Use source audio no longer than 8 minutes. Select Edit and mark the start and end of the intended range on the waveform. Review Lyrics and Style, then generate.
The original is not overwritten. Listen through the new result, especially just before and after the selected range. Edit targets that section but does not promise sample-exact preservation everywhere else. It is different from correcting notes in an editable melody score. See the ACE-Step Cover and Edit instructions.

4. Separate and mix only when separate parts will help
If the performance is right but the balance needs work, open Get stems from the Library song menu. Vocal and Instrumental costs 5 Credits per song; All Detected Stems costs 60. Both require a paid plan. The returned parts depend on detection and may contain bleed or artifacts.
Listen before moving into ACE Studio or another DAW. Export the available parts and import them into a separate session; ACE Music has no automatic Open in ACE Studio transfer. Keep the full mix as a reference. Adjust the few parts responsible for the problem, then compare at a similar listening level.
Stems are separated audio, not the original recording session or automatically editable instrument performances.
5. Compare the final file instead of the format label
Download the available audio and listen to the exact file you will use. ACE-Step supports MP3, not WAV. For other supported models, an offered WAV may be an original service file or a conversion from MP3, depending on availability. Converting the container does not restore lost detail.
A louder file can appear more impressive without being clearer. Compare the same passage at similar playback levels, then hear the whole song for unintended changes. Save the original, chosen revision, and a note of what improved.