How to Create a Song in Musicfy and Add Your Own Custom Voice
Gary WhittakerHow to Create a Song in Musicfy and Add Your Own Custom Voice
This is the exact workflow I used with WAR COMES Part 2. I generated a full song in Musicfy, chose the version I actually liked for its melody and alternate interpretation, preserved that version, then applied my trained custom voice to the selected track instead of regenerating the entire idea from scratch.
The result is useful because it separates two creative decisions: Does the generated song deserve to survive? And then: does my own voice identity make that song stronger?
Hear the workflow before reading it
The first file is the Musicfy generation I chose to keep. The second is the same development branch after I added my trained custom vocal identity. Keep both. The before-and-after comparison is the evidence.
Red Horse War
I kept this version because I liked the melody and the way Musicfy reinterpreted WAR COMES Part 2 as a softer, complementary version rather than a copy of the Suno benchmark.
Red Horse War — Custom Voice Version
After selecting the generation, I used my trained Musicfy custom voice to give the track a more consistent JR vocal identity without discarding the melody and interpretation that made the generation worth keeping.
Where this fits in Find Your Sound
Find Your Sound
This is a production and identity workflow: generate, select, preserve, apply a reusable voice, compare, then decide what advances.
Apply → Build
Apply because you finish with a before/after voice test. Build because the trained voice becomes a reusable production asset across future songs.
One generation worth preserving
You should already have a Musicfy song whose melody, arrangement or emotional interpretation gives you a reason to continue.
A song plus a documented vocal-identity decision
You finish with the original generation, the custom-voice version and a reasoned decision about which elements to preserve.
The process: create the song first, then add your voice
Build the song in Create a Song
Start with the song itself. In this project I kept the original WAR COMES Part 2 lyric intact and used Musicfy-focused section metadata plus a separate Style of Music description. The goal was to give Musicfy enough direction to make a coherent record without rewriting the source lyric.
Generate versions and listen for a reason to keep one
Do not select a version because it is merely complete. Listen for a melody, arrangement, emotional reading, groove or production idea that gives you something worth protecting. Red Horse War earned its place because the melody and softer interpretation opened a complementary direction for WAR COMES.
Preserve the winning generation before changing it
Save the original file. Name it clearly. Do not overwrite it with the custom-voice version. You need the untouched generation to judge whether the voice step actually improved the song.
Train or select your own custom voice
Your custom voice is a separate production asset. Build it from clean, permission-safe source material, test it independently, and keep a record of the dataset and model version you used. Musicfy currently supports trained custom voices, and its current Pro and Studio song-generation plans include Voice Targeting for generating songs in your trained voice.
Apply the custom voice to the generation you chose
This is the point where song selection and voice identity meet. I took the Red Horse War generation I had already chosen and applied my trained custom vocal identity to that development branch. The composition had already earned the right to survive; the new question was whether my voice made it feel more connected to the larger WAR COMES identity.
If you are creating a new song on a Musicfy plan with Voice Targeting, Musicfy can target a trained voice during song generation. If you already have audio whose performance or composition you want to preserve, use the appropriate Musicfy voice workflow rather than rebuilding the entire song only to change the voice.
Compare before and after without letting familiarity decide
Listen to the original generation and the custom-voice version back to back. Judge the voice separately from the song. A familiar voice can feel “better” simply because it is familiar, even when phrasing, pitch or articulation became weaker.
What to compare
| Question | What to listen for |
|---|---|
| Song identity | Did the melody, groove and emotional interpretation that made the generation worth keeping survive the voice step? |
| Vocal identity | Does the result sound more consistently connected to your intended artist or project identity? |
| Cadence | Did the custom voice preserve the phrasing, accents, pauses and rhythmic attack? |
| Pitch stability | Did notes remain natural, or did the conversion introduce unstable or artificial pitch movement? |
| Intelligibility | Can you still understand the lyric clearly, especially in dense sections? |
| Emotion | Did the voice strengthen the feeling of the performance or flatten it? |
| Artifacts | Listen for metallic edges, smeared consonants, doubled character, breath artifacts or unnatural transitions. |
| Value added | Is the custom-voice version meaningfully better for the project, or merely different? |
What this WAR COMES test taught me
The important result is not that Musicfy can generate a song and then put a custom voice on it. The useful part is the separation of decisions.
First, Red Horse War had to earn its place as a song. I liked its melody and its softer reinterpretation of WAR COMES. It was complementary to the harder Suno benchmark rather than a replacement for it.
Then the custom voice became a second creative decision. That let me test whether I could preserve the Musicfy interpretation while bringing the vocal identity closer to the wider WAR COMES project.
This is exactly the kind of workflow I want creators to understand: generate → select → preserve → transform → compare → document. Do not keep regenerating the entire song when the part you actually want to change is the voice.
When this workflow makes sense
Use it when the song works but the singer does not
You found a melody, arrangement or performance worth keeping, but the generated vocal identity is not the identity you want to build around.
Use it when you want consistency across releases
A trained custom voice can become part of a repeatable artist or project system when it is tested and documented carefully.
Do not use it to avoid diagnosing the song
A custom voice will not fix a weak hook, bad lyric fit, poor arrangement or production problem that exists underneath the vocal identity.
Do not confuse technical access with rights clearance
Keep source rights, voice/model permission, plan rights, generated output and release decisions separate.
Build one before-and-after custom-voice test.
Complete the lesson only after you have:
- one Musicfy song generation you genuinely want to keep;
- the untouched source file saved;
- one trained custom voice you have independently tested;
- one version of the selected song using that voice;
- a written comparison of song identity, vocal identity, cadence, intelligibility, emotion and artifacts;
- a clear decision: keep the original, keep the custom-voice version, keep both for different jobs, or revise one bounded problem.
Success criteria: you can explain what changed, why you changed it and why the chosen version advances. “I liked it more” is a starting observation, not the full production decision.
Musicfy Creator Hub → Create a Song → Custom Voice Training → Voice Transformation → Complete Track Workflow → Rights & Release →
Current Musicfy note: Musicfy’s live product currently describes full-song generation from a mood, genre or idea, supports user-supplied lyrics, trained custom voices and Voice Targeting on Pro/Studio plans. Features and plan access can change, so verify the current interface before depending on a specific control. Partner disclosure: Jack Righteous has an ongoing content and affiliate relationship with Musicfy. The workflow and conclusions here come from JR project testing.
