How to Remix a Song with YuE2: AI Covers That Keep the Melody
Remixing with YuE2 is the closest thing Song Creator Pro has to hiring a session band: give it a song, describe a new sound, and it re-performs the track in that style with the melody intact. It's one of the two headline features of YuE2 that are supported in Song Creator Pro alongside text-to-song generation.
This guide walks through the whole workflow: what YuE2 actually does with your song, every step in the Remix tab, how to write the style prompt, and how to change the words without breaking the melody.
If you're looking for the ACE-Step remix workflow (Source Influence, Melody Retention, reference audio), refer to this guide: How to Remix a Song or Create a Song Cover using AI.
What YuE2 Remixing Actually Does
Most AI remix tools transform the source audio directly, morphing the waveform toward a new style. YuE2 takes a different route, and it explains everything about how you use it:
- Transcribe. The source song is transcribed into a symbolic melody score plus its lyrics. This is the same kind of score YuE2 writes when it plans an original song, except here it's learned from your audio.
- Re-perform. YuE2 then generates a brand-new recording of that score, following your style description for the genre, instruments, and vocal character.
The consequence: the sung melody survives the trip, because it's written down in the score, while the arrangement is rebuilt from scratch around your description. A piano ballad becomes a synthwave track that you can still sing along to. That's also why YuE2 needs the lyrics: they're part of the score. No lyrics, no vocal melody to plan around.
In YuE2's official release, the transcription step uses SheetSage2, a companion model from the same team, and these score-based covers are one of YuE2's headline features. For more on the model itself, see what YuE2 is and how it works.
Everything runs locally on your PC, like all Song Creator Pro generation.
Turn any song on your drive into a cover that keeps its melody. Song Creator Pro runs YuE2 remixing on your own PC with an 8 GB NVIDIA GPU.
One-Time Setup: Download the Cover Tools
The transcription models that power remixing are a separate download from YuE2 itself. If you see "The cover tools are not downloaded yet" under the Source Audio field, open the Model picker at the top of the panel and download them from there. You only do this once.
Step 1: Open Remix and Select YuE2
Click the Remix tab at the top of the interface, then make sure the Model selector in the left panel says YuE2. The Remix tab serves both models, and the controls change depending on which one is active: with YuE2 selected you get the Description, Source Audio, and generation settings described below.
Step 2: Upload Your Source Audio
Click the Source Audio field and choose the song you want to remix. It can be a track you generated in Song Creator Pro, a recording of your own, or any song you have on disk.
Songs with a clear lead vocal transcribe best. If the vocal is buried in the mix or heavily processed, transcription still works but expect more errors to fix in the next step.
Step 3: Analyze the Song
Click Analyze next to the Source Audio field. Song Creator Pro transcribes the song, and the lyrics appear in the Lyrics panel, organized into labeled sections like [Verse] and [Chorus]. Instrumental passages get their own labels too, such as [Instrumental Break].
This is the step that saves you the tedious part of YuE2's official cover workflow. Running the model standalone, you'd transcribe the audio yourself with a separate tool and review the raw score by hand. Here it's one click.
Step 4: Review and Edit the Lyrics
Auto-transcription is good, but it isn't perfect, and this step matters more than any other: whatever sits in the lyrics field is what gets sung. A misheard word becomes a sung misheard word.
Go through the transcription and check:
- Misheard words. Listen to the original if you're unsure of a line.
- Section labels. Sections should be in square brackets on their own lines, separated by blank lines, and should match the actual order of the song.
- Stray text. The same rule as all YuE2 prompting applies: nothing in the lyrics field except section tags and words that get sung.
Keep the lyrics matched to the source song. YuE2 plans the remix around the melody it transcribed, and the lyrics need to line up with that melody. This is also where you can rewrite the words entirely; there's a right way to do that, covered in the "Changing the Words" section below.
Step 5: Describe the Destination
The Description field is your style prompt, and it works exactly like a YuE2 style prompt for an original song: comma-separated tags covering genre, instruments, mood, vocal character, language, and tempo.
The one remix-specific rule: describe the destination, not the source. You don't need to tell YuE2 what the original sounds like; that's what the audio is for. Spend every word on where you want the remix to land.
Example destination prompts:
orchestral, cinematic, lush strings, brass swells, expressive male vocal, dramatic build
lo-fi hip hop, dusty drums, jazzy Rhodes piano, laid-back, warm female vocal, nostalgic
80s synthwave, analog pads, gated reverb drums, driving bassline, bright male vocal
The distance between the source genre and your description sets how radical the remix feels. A soul ballad described as "acoustic folk, fingerpicked guitar, intimate male vocal" is a faithful cover. The same ballad described as "cyber metal, distorted guitars, pounding drums" is a reinvention that happens to share a melody.
Step 6: Settings
Four controls sit under the Source Audio field:
- Seed: leave at -1 for a random seed each run. Set a specific number to make results reproducible, useful when you want to tweak only the description and compare fairly.
- Songs: how many candidates to generate in one batch. Set it to 2-4; remixing is exploratory and the first take is rarely the best.
- Quality: leave on Best unless you're doing quick throwaway experiments.
- Format: MP3, or switch to a lossless format if this take might be a keeper.
Step 7: Generate and Iterate
Hit Generate. A three-minute remix takes a few minutes, and each candidate lands in the Results panel with its own Retake, Remix, Revise, and Layer buttons.
Two iteration patterns work well:
- Retake regenerates with the same inputs and a new seed. Use it when the direction is right but the take isn't.
- Chained remixes: click Remix on a result to feed it back in as new source audio, change the description, and generate again. Each pass drifts further from the original, which is a feature: it's how you find versions of a song nobody would have planned.
As with everything YuE2, best-of-N is the habit that pays: generate a few, keep the winner, discard the rest. Local generation makes extra candidates free.
Changing the Words: Covers with New Lyrics
The lyrics field is fully editable, which means YuE2 remixing doubles as a cover machine: same melody, your words. Rewrites, parodies, and translations all work, with one rule that comes straight from the developers' cover documentation: match the phrasing and syllable counts of the original lines.
The transcribed melody expects syllables where the original song put them. If the original line carries nine syllables and your replacement carries fourteen, the vocal gets rushed and unnatural. Line by line:
- Keep the section structure identical. Don't add or remove verses; replace their contents.
- Count the syllables in the original line and aim for the same count in yours.
- Keep the natural stresses in similar places. A line that reads well isn't always a line that sings well on this melody.
For translations, this means adapting rather than translating literally. A line that scans naturally in English usually needs restructuring to sing naturally in French or Japanese on the same notes. And match the language across both fields: if you translate the lyrics to Japanese, say "Japanese" in your description. YuE2 sings in English, Mandarin, and Japanese.
YuE2 or ACE-Step for This Remix?
Song Creator Pro has two remix engines, and they're genuinely different tools:
| You want | Use |
|---|---|
| A genre swap where the sung melody stays clearly recognizable | YuE2 |
| A cover with rewritten or translated lyrics | YuE2 |
| A dial for exactly how close the remix stays to the source | ACE-Step 1.5 |
| To blend in the timbre of a second reference track | ACE-Step 1.5 |
| An instrumental remix of a vocal track | ACE-Step 1.5 |
| Vocals in a language beyond English, Mandarin, Japanese | ACE-Step 1.5 |
The short version: YuE2 remixes are covers, rebuilt from a score with the melody guaranteed. ACE-Step remixes are transformations, sculpted from the audio with continuous control. Both are one click from any result in Song Creator Pro, so trying both on the same song costs nothing.
Related Guides
- What Is YuE2?: who made the model, how it works, and how it compares to YuE and ACE-Step
- YuE2 Prompting Guide: style tags, lyric formatting, and planning modes for original songs
- How to Remix a Song with AI (ACE-Step): the Source Influence and Melody Retention workflow
- Extract: pull vocals or instruments out of a song
- Getting Started Guide: the basics of generating songs from scratch
Ready to remix? Two AI models including YuE2, unlimited local generation on your Windows PC.
Try it free on the Microsoft Store
One-time purchase · Lifetime updates · Commercial license
$49.99 on the Microsoft Store, free trial included · or $44.99 on itch.io
Frequently Asked Questions
ACE-Step transforms the source audio directly and gives you dials like Source Influence and Melody Retention to control how far it drifts. YuE2 works like a musician covering a song: it transcribes the source into a melody score and lyrics, then re-performs that score in the style you describe. The result keeps the sung melody clearly recognizable while rebuilding the entire arrangement. Use YuE2 for melody-faithful genre swaps and covers; use ACE-Step when you want continuous control over how close the remix stays.
YuE2 plans every song as a symbolic score before rendering audio, and for a remix that plan is built around the source song's melody and words. The lyrics anchor the vocal line so the new version sings the same song. Song Creator Pro auto-transcribes the lyrics from your source audio, and you can review and edit them before generating.
Yes, the transcribed lyrics are fully editable, which is how you make covers with new words or translations. The catch is that the lyrics still need to fit the source song's melody: keep the same section structure and match the phrasing and syllable counts of the original lines. A line with a very different syllable count forces rushed or stretched delivery.
Songs with a clear lead vocal melody transcribe most accurately, which makes for the best remixes. Vocals in English, Mandarin, or Japanese work best since those are the languages YuE2 sings. Very dense mixes, heavy vocal processing, or buried vocals make transcription harder, so expect to do more cleanup in the lyrics field for those.
The melody is preserved by design, so closeness is mostly about your style prompt and iteration. A style prompt near the original genre produces a faithful cover; a distant one produces a full reinvention. Generate 2-4 candidates per prompt, keep the best, and use Retake or chained remixes to iterate. If you want continuous control over how much of the source survives, ACE-Step's remix with its Source Influence dial is the better tool.